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_MultTransposeAdd, MAT_Solve, MAT_Solves, MAT_SolveAdd, MAT_SolveTranspose, MAT_MatSolve, MAT_MatTrSolve;
19: PetscLogEvent MAT_SolveTransposeAdd, MAT_SOR, MAT_ForwardSolve, MAT_BackwardSolve, MAT_LUFactor, MAT_LUFactorSymbolic;
20: PetscLogEvent MAT_LUFactorNumeric, MAT_CholeskyFactor, MAT_CholeskyFactorSymbolic, MAT_CholeskyFactorNumeric, MAT_ILUFactor;
21: PetscLogEvent MAT_ILUFactorSymbolic, MAT_ICCFactorSymbolic, MAT_Copy, MAT_Convert, MAT_Scale, MAT_AssemblyBegin;
22: PetscLogEvent MAT_QRFactorNumeric, MAT_QRFactorSymbolic, MAT_QRFactor;
23: PetscLogEvent MAT_AssemblyEnd, MAT_SetValues, MAT_GetValues, MAT_GetRow, MAT_GetRowIJ, MAT_CreateSubMats, MAT_GetOrdering, MAT_RedundantMat, MAT_GetSeqNonzeroStructure;
24: PetscLogEvent MAT_IncreaseOverlap, MAT_Partitioning, MAT_PartitioningND, MAT_Coarsen, MAT_ZeroEntries, MAT_Load, MAT_View, MAT_AXPY, MAT_FDColoringCreate;
25: PetscLogEvent MAT_FDColoringSetUp, MAT_FDColoringApply, MAT_Transpose, MAT_FDColoringFunction, MAT_CreateSubMat;
26: PetscLogEvent MAT_TransposeColoringCreate;
27: PetscLogEvent MAT_MatMult, MAT_MatMultSymbolic, MAT_MatMultNumeric;
28: PetscLogEvent MAT_PtAP, MAT_PtAPSymbolic, MAT_PtAPNumeric, MAT_RARt, MAT_RARtSymbolic, MAT_RARtNumeric;
29: PetscLogEvent MAT_MatTransposeMult, MAT_MatTransposeMultSymbolic, MAT_MatTransposeMultNumeric;
30: PetscLogEvent MAT_TransposeMatMult, MAT_TransposeMatMultSymbolic, MAT_TransposeMatMultNumeric;
31: PetscLogEvent MAT_MatMatMult, MAT_MatMatMultSymbolic, MAT_MatMatMultNumeric;
32: PetscLogEvent MAT_MultHermitianTranspose, MAT_MultHermitianTransposeAdd;
33: PetscLogEvent MAT_Getsymtransreduced, MAT_GetBrowsOfAcols;
34: PetscLogEvent MAT_GetBrowsOfAocols, MAT_Getlocalmat, MAT_Getlocalmatcondensed, MAT_Seqstompi, MAT_Seqstompinum, MAT_Seqstompisym;
35: PetscLogEvent MAT_GetMultiProcBlock;
36: PetscLogEvent MAT_CUSPARSECopyToGPU, MAT_CUSPARSECopyFromGPU, MAT_CUSPARSEGenerateTranspose, MAT_CUSPARSESolveAnalysis;
37: PetscLogEvent MAT_HIPSPARSECopyToGPU, MAT_HIPSPARSECopyFromGPU, MAT_HIPSPARSEGenerateTranspose, MAT_HIPSPARSESolveAnalysis;
38: PetscLogEvent MAT_PreallCOO, MAT_SetVCOO;
39: PetscLogEvent MAT_CreateGraph;
40: PetscLogEvent MAT_SetValuesBatch;
41: PetscLogEvent MAT_ViennaCLCopyToGPU;
42: PetscLogEvent MAT_CUDACopyToGPU, MAT_HIPCopyToGPU;
43: PetscLogEvent MAT_DenseCopyToGPU, MAT_DenseCopyFromGPU;
44: PetscLogEvent MAT_Merge, MAT_Residual, MAT_SetRandom;
45: PetscLogEvent MAT_FactorFactS, MAT_FactorInvS;
46: PetscLogEvent MATCOLORING_Apply, MATCOLORING_Comm, MATCOLORING_Local, MATCOLORING_ISCreate, MATCOLORING_SetUp, MATCOLORING_Weights;
47: PetscLogEvent MAT_H2Opus_Build, MAT_H2Opus_Compress, MAT_H2Opus_Orthog, MAT_H2Opus_LR;
49: const char *const MatFactorTypes[] = {"NONE", "LU", "CHOLESKY", "ILU", "ICC", "ILUDT", "QR", "MatFactorType", "MAT_FACTOR_", NULL};
51: /*@
52: MatSetRandom - Sets all components of a matrix to random numbers.
54: Logically Collective
56: Input Parameters:
57: + x - the matrix
58: - rctx - the `PetscRandom` object, formed by `PetscRandomCreate()`, or `NULL` and
59: it will create one internally.
61: Example:
62: .vb
63: PetscRandomCreate(PETSC_COMM_WORLD,&rctx);
64: MatSetRandom(x,rctx);
65: PetscRandomDestroy(rctx);
66: .ve
68: Level: intermediate
70: Notes:
71: For sparse matrices that have been preallocated but not been assembled, it randomly selects appropriate locations,
73: for sparse matrices that already have nonzero locations, it fills the locations with random numbers.
75: It generates an error if used on unassembled sparse matrices that have not been preallocated.
77: .seealso: [](ch_matrices), `Mat`, `PetscRandom`, `PetscRandomCreate()`, `MatZeroEntries()`, `MatSetValues()`, `PetscRandomDestroy()`
78: @*/
79: PetscErrorCode MatSetRandom(Mat x, PetscRandom rctx)
80: {
81: PetscRandom randObj = NULL;
83: PetscFunctionBegin;
87: MatCheckPreallocated(x, 1);
89: if (!rctx) {
90: MPI_Comm comm;
91: PetscCall(PetscObjectGetComm((PetscObject)x, &comm));
92: PetscCall(PetscRandomCreate(comm, &randObj));
93: PetscCall(PetscRandomSetType(randObj, x->defaultrandtype));
94: PetscCall(PetscRandomSetFromOptions(randObj));
95: rctx = randObj;
96: }
97: PetscCall(PetscLogEventBegin(MAT_SetRandom, x, rctx, 0, 0));
98: PetscUseTypeMethod(x, setrandom, rctx);
99: PetscCall(PetscLogEventEnd(MAT_SetRandom, x, rctx, 0, 0));
101: PetscCall(MatAssemblyBegin(x, MAT_FINAL_ASSEMBLY));
102: PetscCall(MatAssemblyEnd(x, MAT_FINAL_ASSEMBLY));
103: PetscCall(PetscRandomDestroy(&randObj));
104: PetscFunctionReturn(PETSC_SUCCESS);
105: }
107: /*@
108: MatFactorGetErrorZeroPivot - returns the pivot value that was determined to be zero and the row it occurred in
110: Logically Collective
112: Input Parameter:
113: . mat - the factored matrix
115: Output Parameters:
116: + pivot - the pivot value computed
117: - row - the row that the zero pivot occurred. This row value must be interpreted carefully due to row reorderings and which processes
118: the share the matrix
120: Level: advanced
122: Notes:
123: This routine does not work for factorizations done with external packages.
125: This routine should only be called if `MatGetFactorError()` returns a value of `MAT_FACTOR_NUMERIC_ZEROPIVOT`
127: This can also be called on non-factored matrices that come from, for example, matrices used in SOR.
129: .seealso: [](ch_matrices), `Mat`, `MatZeroEntries()`, `MatFactor()`, `MatGetFactor()`,
130: `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`, `MatFactorClearError()`,
131: `MAT_FACTOR_NUMERIC_ZEROPIVOT`
132: @*/
133: PetscErrorCode MatFactorGetErrorZeroPivot(Mat mat, PetscReal *pivot, PetscInt *row)
134: {
135: PetscFunctionBegin;
137: PetscAssertPointer(pivot, 2);
138: PetscAssertPointer(row, 3);
139: *pivot = mat->factorerror_zeropivot_value;
140: *row = mat->factorerror_zeropivot_row;
141: PetscFunctionReturn(PETSC_SUCCESS);
142: }
144: /*@
145: MatFactorGetError - gets the error code from a factorization
147: Logically Collective
149: Input Parameter:
150: . mat - the factored matrix
152: Output Parameter:
153: . err - the error code
155: Level: advanced
157: Note:
158: This can also be called on non-factored matrices that come from, for example, matrices used in SOR.
160: .seealso: [](ch_matrices), `Mat`, `MatZeroEntries()`, `MatFactor()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`,
161: `MatFactorClearError()`, `MatFactorGetErrorZeroPivot()`, `MatFactorError`
162: @*/
163: PetscErrorCode MatFactorGetError(Mat mat, MatFactorError *err)
164: {
165: PetscFunctionBegin;
167: PetscAssertPointer(err, 2);
168: *err = mat->factorerrortype;
169: PetscFunctionReturn(PETSC_SUCCESS);
170: }
172: /*@
173: MatFactorClearError - clears the error code in a factorization
175: Logically Collective
177: Input Parameter:
178: . mat - the factored matrix
180: Level: developer
182: Note:
183: This can also be called on non-factored matrices that come from, for example, matrices used in SOR.
185: .seealso: [](ch_matrices), `Mat`, `MatZeroEntries()`, `MatFactor()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`, `MatFactorGetError()`, `MatFactorGetErrorZeroPivot()`,
186: `MatGetErrorCode()`, `MatFactorError`
187: @*/
188: PetscErrorCode MatFactorClearError(Mat mat)
189: {
190: PetscFunctionBegin;
192: mat->factorerrortype = MAT_FACTOR_NOERROR;
193: mat->factorerror_zeropivot_value = 0.0;
194: mat->factorerror_zeropivot_row = 0;
195: PetscFunctionReturn(PETSC_SUCCESS);
196: }
198: PetscErrorCode MatFindNonzeroRowsOrCols_Basic(Mat mat, PetscBool cols, PetscReal tol, IS *nonzero)
199: {
200: Vec r, l;
201: const PetscScalar *al;
202: PetscInt i, nz, gnz, N, n, st;
204: PetscFunctionBegin;
205: PetscCall(MatCreateVecs(mat, &r, &l));
206: if (!cols) { /* nonzero rows */
207: PetscCall(MatGetOwnershipRange(mat, &st, NULL));
208: PetscCall(MatGetSize(mat, &N, NULL));
209: PetscCall(MatGetLocalSize(mat, &n, NULL));
210: PetscCall(VecSet(l, 0.0));
211: PetscCall(VecSetRandom(r, NULL));
212: PetscCall(MatMult(mat, r, l));
213: PetscCall(VecGetArrayRead(l, &al));
214: } else { /* nonzero columns */
215: PetscCall(MatGetOwnershipRangeColumn(mat, &st, NULL));
216: PetscCall(MatGetSize(mat, NULL, &N));
217: PetscCall(MatGetLocalSize(mat, NULL, &n));
218: PetscCall(VecSet(r, 0.0));
219: PetscCall(VecSetRandom(l, NULL));
220: PetscCall(MatMultTranspose(mat, l, r));
221: PetscCall(VecGetArrayRead(r, &al));
222: }
223: if (tol <= 0.0) {
224: for (i = 0, nz = 0; i < n; i++)
225: if (al[i] != 0.0) nz++;
226: } else {
227: for (i = 0, nz = 0; i < n; i++)
228: if (PetscAbsScalar(al[i]) > tol) nz++;
229: }
230: PetscCallMPI(MPIU_Allreduce(&nz, &gnz, 1, MPIU_INT, MPI_SUM, PetscObjectComm((PetscObject)mat)));
231: if (gnz != N) {
232: PetscInt *nzr;
233: PetscCall(PetscMalloc1(nz, &nzr));
234: if (nz) {
235: if (tol < 0) {
236: for (i = 0, nz = 0; i < n; i++)
237: if (al[i] != 0.0) nzr[nz++] = i + st;
238: } else {
239: for (i = 0, nz = 0; i < n; i++)
240: if (PetscAbsScalar(al[i]) > tol) nzr[nz++] = i + st;
241: }
242: }
243: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)mat), nz, nzr, PETSC_OWN_POINTER, nonzero));
244: } else *nonzero = NULL;
245: if (!cols) { /* nonzero rows */
246: PetscCall(VecRestoreArrayRead(l, &al));
247: } else {
248: PetscCall(VecRestoreArrayRead(r, &al));
249: }
250: PetscCall(VecDestroy(&l));
251: PetscCall(VecDestroy(&r));
252: PetscFunctionReturn(PETSC_SUCCESS);
253: }
255: /*@
256: MatFindNonzeroRows - Locate all rows that are not completely zero in the matrix
258: Input Parameter:
259: . mat - the matrix
261: Output Parameter:
262: . keptrows - the rows that are not completely zero
264: Level: intermediate
266: Note:
267: `keptrows` is set to `NULL` if all rows are nonzero.
269: Developer Note:
270: If `keptrows` is not `NULL`, it must be sorted.
272: .seealso: [](ch_matrices), `Mat`, `MatFindZeroRows()`
273: @*/
274: PetscErrorCode MatFindNonzeroRows(Mat mat, IS *keptrows)
275: {
276: PetscFunctionBegin;
279: PetscAssertPointer(keptrows, 2);
280: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
281: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
282: if (mat->ops->findnonzerorows) PetscUseTypeMethod(mat, findnonzerorows, keptrows);
283: else PetscCall(MatFindNonzeroRowsOrCols_Basic(mat, PETSC_FALSE, 0.0, keptrows));
284: if (keptrows && *keptrows) PetscCall(ISSetInfo(*keptrows, IS_SORTED, IS_GLOBAL, PETSC_FALSE, PETSC_TRUE));
285: PetscFunctionReturn(PETSC_SUCCESS);
286: }
288: /*@
289: MatFindZeroRows - Locate all rows that are completely zero in the matrix
291: Input Parameter:
292: . mat - the matrix
294: Output Parameter:
295: . zerorows - the rows that are completely zero
297: Level: intermediate
299: Note:
300: `zerorows` is set to `NULL` if no rows are zero.
302: .seealso: [](ch_matrices), `Mat`, `MatFindNonzeroRows()`
303: @*/
304: PetscErrorCode MatFindZeroRows(Mat mat, IS *zerorows)
305: {
306: IS keptrows;
307: PetscInt m, n;
309: PetscFunctionBegin;
312: PetscAssertPointer(zerorows, 2);
313: PetscCall(MatFindNonzeroRows(mat, &keptrows));
314: /* MatFindNonzeroRows sets keptrows to NULL if there are no zero rows.
315: In keeping with this convention, we set zerorows to NULL if there are no zero
316: rows. */
317: if (keptrows == NULL) {
318: *zerorows = NULL;
319: } else {
320: PetscCall(MatGetOwnershipRange(mat, &m, &n));
321: PetscCall(ISComplement(keptrows, m, n, zerorows));
322: PetscCall(ISDestroy(&keptrows));
323: }
324: PetscFunctionReturn(PETSC_SUCCESS);
325: }
327: /*@
328: MatGetDiagonalBlock - Returns the part of the matrix associated with the on-process coupling
330: Not Collective
332: Input Parameter:
333: . A - the matrix
335: Output Parameter:
336: . a - the diagonal part (which is a SEQUENTIAL matrix)
338: Level: advanced
340: Notes:
341: See `MatCreateAIJ()` for more information on the "diagonal part" of the matrix.
343: Use caution, as the reference count on the returned matrix is not incremented and it is used as part of `A`'s normal operation.
345: .seealso: [](ch_matrices), `Mat`, `MatCreateAIJ()`, `MATAIJ`, `MATBAIJ`, `MATSBAIJ`
346: @*/
347: PetscErrorCode MatGetDiagonalBlock(Mat A, Mat *a)
348: {
349: PetscFunctionBegin;
352: PetscAssertPointer(a, 2);
353: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
354: if (A->ops->getdiagonalblock) PetscUseTypeMethod(A, getdiagonalblock, a);
355: else {
356: PetscMPIInt size;
358: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A), &size));
359: PetscCheck(size == 1, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Not for parallel matrix type %s", ((PetscObject)A)->type_name);
360: *a = A;
361: }
362: PetscFunctionReturn(PETSC_SUCCESS);
363: }
365: /*@
366: MatGetTrace - Gets the trace of a matrix. The sum of the diagonal entries.
368: Collective
370: Input Parameter:
371: . mat - the matrix
373: Output Parameter:
374: . trace - the sum of the diagonal entries
376: Level: advanced
378: .seealso: [](ch_matrices), `Mat`
379: @*/
380: PetscErrorCode MatGetTrace(Mat mat, PetscScalar *trace)
381: {
382: Vec diag;
384: PetscFunctionBegin;
386: PetscAssertPointer(trace, 2);
387: PetscCall(MatCreateVecs(mat, &diag, NULL));
388: PetscCall(MatGetDiagonal(mat, diag));
389: PetscCall(VecSum(diag, trace));
390: PetscCall(VecDestroy(&diag));
391: PetscFunctionReturn(PETSC_SUCCESS);
392: }
394: /*@
395: MatRealPart - Zeros out the imaginary part of the matrix
397: Logically Collective
399: Input Parameter:
400: . mat - the matrix
402: Level: advanced
404: .seealso: [](ch_matrices), `Mat`, `MatImaginaryPart()`
405: @*/
406: PetscErrorCode MatRealPart(Mat mat)
407: {
408: PetscFunctionBegin;
411: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
412: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
413: MatCheckPreallocated(mat, 1);
414: PetscUseTypeMethod(mat, realpart);
415: PetscFunctionReturn(PETSC_SUCCESS);
416: }
418: /*@C
419: MatGetGhosts - Get the global indices of all ghost nodes defined by the sparse matrix
421: Collective
423: Input Parameter:
424: . mat - the matrix
426: Output Parameters:
427: + nghosts - number of ghosts (for `MATBAIJ` and `MATSBAIJ` matrices there is one ghost for each matrix block)
428: - ghosts - the global indices of the ghost points
430: Level: advanced
432: Note:
433: `nghosts` and `ghosts` are suitable to pass into `VecCreateGhost()` or `VecCreateGhostBlock()`
435: .seealso: [](ch_matrices), `Mat`, `VecCreateGhost()`, `VecCreateGhostBlock()`
436: @*/
437: PetscErrorCode MatGetGhosts(Mat mat, PetscInt *nghosts, const PetscInt *ghosts[])
438: {
439: PetscFunctionBegin;
442: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
443: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
444: if (mat->ops->getghosts) PetscUseTypeMethod(mat, getghosts, nghosts, ghosts);
445: else {
446: if (nghosts) *nghosts = 0;
447: if (ghosts) *ghosts = NULL;
448: }
449: PetscFunctionReturn(PETSC_SUCCESS);
450: }
452: /*@
453: MatImaginaryPart - Moves the imaginary part of the matrix to the real part and zeros the imaginary part
455: Logically Collective
457: Input Parameter:
458: . mat - the matrix
460: Level: advanced
462: .seealso: [](ch_matrices), `Mat`, `MatRealPart()`
463: @*/
464: PetscErrorCode MatImaginaryPart(Mat mat)
465: {
466: PetscFunctionBegin;
469: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
470: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
471: MatCheckPreallocated(mat, 1);
472: PetscUseTypeMethod(mat, imaginarypart);
473: PetscFunctionReturn(PETSC_SUCCESS);
474: }
476: /*@
477: MatMissingDiagonal - Determine if sparse matrix is missing a diagonal entry (or block entry for `MATBAIJ` and `MATSBAIJ` matrices) in the nonzero structure
479: Not Collective
481: Input Parameter:
482: . mat - the matrix
484: Output Parameters:
485: + missing - is any diagonal entry missing
486: - dd - first diagonal entry that is missing (optional) on this process
488: Level: advanced
490: Note:
491: This does not return diagonal entries that are in the nonzero structure but happen to have a zero numerical value
493: .seealso: [](ch_matrices), `Mat`
494: @*/
495: PetscErrorCode MatMissingDiagonal(Mat mat, PetscBool *missing, PetscInt *dd)
496: {
497: PetscFunctionBegin;
500: PetscAssertPointer(missing, 2);
501: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix %s", ((PetscObject)mat)->type_name);
502: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
503: PetscUseTypeMethod(mat, missingdiagonal, missing, dd);
504: PetscFunctionReturn(PETSC_SUCCESS);
505: }
507: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
508: /*@C
509: MatGetRow - Gets a row of a matrix. You MUST call `MatRestoreRow()`
510: for each row that you get to ensure that your application does
511: not bleed memory.
513: Not Collective
515: Input Parameters:
516: + mat - the matrix
517: - row - the row to get
519: Output Parameters:
520: + ncols - if not `NULL`, the number of nonzeros in `row`
521: . cols - if not `NULL`, the column numbers
522: - vals - if not `NULL`, the numerical values
524: Level: advanced
526: Notes:
527: This routine is provided for people who need to have direct access
528: to the structure of a matrix. We hope that we provide enough
529: high-level matrix routines that few users will need it.
531: `MatGetRow()` always returns 0-based column indices, regardless of
532: whether the internal representation is 0-based (default) or 1-based.
534: For better efficiency, set `cols` and/or `vals` to `NULL` if you do
535: not wish to extract these quantities.
537: The user can only examine the values extracted with `MatGetRow()`;
538: the values CANNOT be altered. To change the matrix entries, one
539: must use `MatSetValues()`.
541: You can only have one call to `MatGetRow()` outstanding for a particular
542: matrix at a time, per processor. `MatGetRow()` can only obtain rows
543: associated with the given processor, it cannot get rows from the
544: other processors; for that we suggest using `MatCreateSubMatrices()`, then
545: `MatGetRow()` on the submatrix. The row index passed to `MatGetRow()`
546: is in the global number of rows.
548: Use `MatGetRowIJ()` and `MatRestoreRowIJ()` to access all the local indices of the sparse matrix.
550: Use `MatSeqAIJGetArray()` and similar functions to access the numerical values for certain matrix types directly.
552: Fortran Note:
553: The calling sequence is
554: .vb
555: MatGetRow(matrix,row,ncols,cols,values,ierr)
556: Mat matrix (input)
557: PetscInt row (input)
558: PetscInt ncols (output)
559: PetscInt cols(maxcols) (output)
560: PetscScalar values(maxcols) output
561: .ve
562: where maxcols >= maximum nonzeros in any row of the matrix.
564: .seealso: [](ch_matrices), `Mat`, `MatRestoreRow()`, `MatSetValues()`, `MatGetValues()`, `MatCreateSubMatrices()`, `MatGetDiagonal()`, `MatGetRowIJ()`, `MatRestoreRowIJ()`
565: @*/
566: PetscErrorCode MatGetRow(Mat mat, PetscInt row, PetscInt *ncols, const PetscInt *cols[], const PetscScalar *vals[])
567: {
568: PetscInt incols;
570: PetscFunctionBegin;
573: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
574: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
575: MatCheckPreallocated(mat, 1);
576: 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);
577: PetscCall(PetscLogEventBegin(MAT_GetRow, mat, 0, 0, 0));
578: PetscUseTypeMethod(mat, getrow, row, &incols, (PetscInt **)cols, (PetscScalar **)vals);
579: if (ncols) *ncols = incols;
580: PetscCall(PetscLogEventEnd(MAT_GetRow, mat, 0, 0, 0));
581: PetscFunctionReturn(PETSC_SUCCESS);
582: }
584: /*@
585: MatConjugate - replaces the matrix values with their complex conjugates
587: Logically Collective
589: Input Parameter:
590: . mat - the matrix
592: Level: advanced
594: .seealso: [](ch_matrices), `Mat`, `MatRealPart()`, `MatImaginaryPart()`, `VecConjugate()`, `MatTranspose()`
595: @*/
596: PetscErrorCode MatConjugate(Mat mat)
597: {
598: PetscFunctionBegin;
600: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
601: if (PetscDefined(USE_COMPLEX) && mat->hermitian != PETSC_BOOL3_TRUE) {
602: PetscUseTypeMethod(mat, conjugate);
603: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
604: }
605: PetscFunctionReturn(PETSC_SUCCESS);
606: }
608: /*@C
609: MatRestoreRow - Frees any temporary space allocated by `MatGetRow()`.
611: Not Collective
613: Input Parameters:
614: + mat - the matrix
615: . row - the row to get
616: . ncols - the number of nonzeros
617: . cols - the columns of the nonzeros
618: - vals - if nonzero the column values
620: Level: advanced
622: Notes:
623: This routine should be called after you have finished examining the entries.
625: This routine zeros out `ncols`, `cols`, and `vals`. This is to prevent accidental
626: us of the array after it has been restored. If you pass `NULL`, it will
627: not zero the pointers. Use of `cols` or `vals` after `MatRestoreRow()` is invalid.
629: Fortran Note:
630: `MatRestoreRow()` MUST be called after `MatGetRow()`
631: before another call to `MatGetRow()` can be made.
633: .seealso: [](ch_matrices), `Mat`, `MatGetRow()`
634: @*/
635: PetscErrorCode MatRestoreRow(Mat mat, PetscInt row, PetscInt *ncols, const PetscInt *cols[], const PetscScalar *vals[])
636: {
637: PetscFunctionBegin;
639: if (ncols) PetscAssertPointer(ncols, 3);
640: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
641: if (!mat->ops->restorerow) PetscFunctionReturn(PETSC_SUCCESS);
642: PetscUseTypeMethod(mat, restorerow, row, ncols, (PetscInt **)cols, (PetscScalar **)vals);
643: if (ncols) *ncols = 0;
644: if (cols) *cols = NULL;
645: if (vals) *vals = NULL;
646: PetscFunctionReturn(PETSC_SUCCESS);
647: }
649: /*@
650: MatGetRowUpperTriangular - Sets a flag to enable calls to `MatGetRow()` for matrix in `MATSBAIJ` format.
651: You should call `MatRestoreRowUpperTriangular()` after calling` MatGetRow()` and `MatRestoreRow()` to disable the flag.
653: Not Collective
655: Input Parameter:
656: . mat - the matrix
658: Level: advanced
660: Note:
661: 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.
663: .seealso: [](ch_matrices), `Mat`, `MATSBAIJ`, `MatRestoreRowUpperTriangular()`
664: @*/
665: PetscErrorCode MatGetRowUpperTriangular(Mat mat)
666: {
667: PetscFunctionBegin;
670: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
671: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
672: MatCheckPreallocated(mat, 1);
673: if (!mat->ops->getrowuppertriangular) PetscFunctionReturn(PETSC_SUCCESS);
674: PetscUseTypeMethod(mat, getrowuppertriangular);
675: PetscFunctionReturn(PETSC_SUCCESS);
676: }
678: /*@
679: MatRestoreRowUpperTriangular - Disable calls to `MatGetRow()` for matrix in `MATSBAIJ` format.
681: Not Collective
683: Input Parameter:
684: . mat - the matrix
686: Level: advanced
688: Note:
689: This routine should be called after you have finished calls to `MatGetRow()` and `MatRestoreRow()`.
691: .seealso: [](ch_matrices), `Mat`, `MATSBAIJ`, `MatGetRowUpperTriangular()`
692: @*/
693: PetscErrorCode MatRestoreRowUpperTriangular(Mat mat)
694: {
695: PetscFunctionBegin;
698: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
699: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
700: MatCheckPreallocated(mat, 1);
701: if (!mat->ops->restorerowuppertriangular) PetscFunctionReturn(PETSC_SUCCESS);
702: PetscUseTypeMethod(mat, restorerowuppertriangular);
703: PetscFunctionReturn(PETSC_SUCCESS);
704: }
706: /*@
707: MatSetOptionsPrefix - Sets the prefix used for searching for all
708: `Mat` options in the database.
710: Logically Collective
712: Input Parameters:
713: + A - the matrix
714: - prefix - the prefix to prepend to all option names
716: Level: advanced
718: Notes:
719: A hyphen (-) must NOT be given at the beginning of the prefix name.
720: The first character of all runtime options is AUTOMATICALLY the hyphen.
722: This is NOT used for options for the factorization of the matrix. Normally the
723: prefix is automatically passed in from the PC calling the factorization. To set
724: it directly use `MatSetOptionsPrefixFactor()`
726: .seealso: [](ch_matrices), `Mat`, `MatSetFromOptions()`, `MatSetOptionsPrefixFactor()`
727: @*/
728: PetscErrorCode MatSetOptionsPrefix(Mat A, const char prefix[])
729: {
730: PetscFunctionBegin;
732: PetscCall(PetscObjectSetOptionsPrefix((PetscObject)A, prefix));
733: PetscFunctionReturn(PETSC_SUCCESS);
734: }
736: /*@
737: MatSetOptionsPrefixFactor - Sets the prefix used for searching for all matrix factor options in the database for
738: for matrices created with `MatGetFactor()`
740: Logically Collective
742: Input Parameters:
743: + A - the matrix
744: - prefix - the prefix to prepend to all option names for the factored matrix
746: Level: developer
748: Notes:
749: A hyphen (-) must NOT be given at the beginning of the prefix name.
750: The first character of all runtime options is AUTOMATICALLY the hyphen.
752: Normally the prefix is automatically passed in from the `PC` calling the factorization. To set
753: it directly when not using `KSP`/`PC` use `MatSetOptionsPrefixFactor()`
755: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSetFromOptions()`, `MatSetOptionsPrefix()`, `MatAppendOptionsPrefixFactor()`
756: @*/
757: PetscErrorCode MatSetOptionsPrefixFactor(Mat A, const char prefix[])
758: {
759: PetscFunctionBegin;
761: if (prefix) {
762: PetscAssertPointer(prefix, 2);
763: PetscCheck(prefix[0] != '-', PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Options prefix should not begin with a hyphen");
764: if (prefix != A->factorprefix) {
765: PetscCall(PetscFree(A->factorprefix));
766: PetscCall(PetscStrallocpy(prefix, &A->factorprefix));
767: }
768: } else PetscCall(PetscFree(A->factorprefix));
769: PetscFunctionReturn(PETSC_SUCCESS);
770: }
772: /*@
773: MatAppendOptionsPrefixFactor - Appends to the prefix used for searching for all matrix factor options in the database for
774: for matrices created with `MatGetFactor()`
776: Logically Collective
778: Input Parameters:
779: + A - the matrix
780: - prefix - the prefix to prepend to all option names for the factored matrix
782: Level: developer
784: Notes:
785: A hyphen (-) must NOT be given at the beginning of the prefix name.
786: The first character of all runtime options is AUTOMATICALLY the hyphen.
788: Normally the prefix is automatically passed in from the `PC` calling the factorization. To set
789: it directly when not using `KSP`/`PC` use `MatAppendOptionsPrefixFactor()`
791: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `PetscOptionsCreate()`, `PetscOptionsDestroy()`, `PetscObjectSetOptionsPrefix()`, `PetscObjectPrependOptionsPrefix()`,
792: `PetscObjectGetOptionsPrefix()`, `TSAppendOptionsPrefix()`, `SNESAppendOptionsPrefix()`, `KSPAppendOptionsPrefix()`, `MatSetOptionsPrefixFactor()`,
793: `MatSetOptionsPrefix()`
794: @*/
795: PetscErrorCode MatAppendOptionsPrefixFactor(Mat A, const char prefix[])
796: {
797: size_t len1, len2, new_len;
799: PetscFunctionBegin;
801: if (!prefix) PetscFunctionReturn(PETSC_SUCCESS);
802: if (!A->factorprefix) {
803: PetscCall(MatSetOptionsPrefixFactor(A, prefix));
804: PetscFunctionReturn(PETSC_SUCCESS);
805: }
806: PetscCheck(prefix[0] != '-', PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Options prefix should not begin with a hyphen");
808: PetscCall(PetscStrlen(A->factorprefix, &len1));
809: PetscCall(PetscStrlen(prefix, &len2));
810: new_len = len1 + len2 + 1;
811: PetscCall(PetscRealloc(new_len * sizeof(*A->factorprefix), &A->factorprefix));
812: PetscCall(PetscStrncpy(A->factorprefix + len1, prefix, len2 + 1));
813: PetscFunctionReturn(PETSC_SUCCESS);
814: }
816: /*@
817: MatAppendOptionsPrefix - Appends to the prefix used for searching for all
818: matrix options in the database.
820: Logically Collective
822: Input Parameters:
823: + A - the matrix
824: - prefix - the prefix to prepend to all option names
826: Level: advanced
828: Note:
829: A hyphen (-) must NOT be given at the beginning of the prefix name.
830: The first character of all runtime options is AUTOMATICALLY the hyphen.
832: .seealso: [](ch_matrices), `Mat`, `MatGetOptionsPrefix()`, `MatAppendOptionsPrefixFactor()`, `MatSetOptionsPrefix()`
833: @*/
834: PetscErrorCode MatAppendOptionsPrefix(Mat A, const char prefix[])
835: {
836: PetscFunctionBegin;
838: PetscCall(PetscObjectAppendOptionsPrefix((PetscObject)A, prefix));
839: PetscFunctionReturn(PETSC_SUCCESS);
840: }
842: /*@
843: MatGetOptionsPrefix - Gets the prefix used for searching for all
844: matrix options in the database.
846: Not Collective
848: Input Parameter:
849: . A - the matrix
851: Output Parameter:
852: . prefix - pointer to the prefix string used
854: Level: advanced
856: Fortran Note:
857: The user should pass in a string `prefix` of
858: sufficient length to hold the prefix.
860: .seealso: [](ch_matrices), `Mat`, `MatAppendOptionsPrefix()`, `MatSetOptionsPrefix()`, `MatAppendOptionsPrefixFactor()`, `MatSetOptionsPrefixFactor()`
861: @*/
862: PetscErrorCode MatGetOptionsPrefix(Mat A, const char *prefix[])
863: {
864: PetscFunctionBegin;
866: PetscAssertPointer(prefix, 2);
867: PetscCall(PetscObjectGetOptionsPrefix((PetscObject)A, prefix));
868: PetscFunctionReturn(PETSC_SUCCESS);
869: }
871: /*@
872: MatGetState - Gets the state of a `Mat`. Same value as returned by `PetscObjectStateGet()`
874: Not Collective
876: Input Parameter:
877: . A - the matrix
879: Output Parameter:
880: . state - the object state
882: Level: advanced
884: Note:
885: Object state is an integer which gets increased every time
886: the object is changed. By saving and later querying the object state
887: one can determine whether information about the object is still current.
889: See `MatGetNonzeroState()` to determine if the nonzero structure of the matrix has changed.
891: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `PetscObjectStateGet()`, `MatGetNonzeroState()`
892: @*/
893: PetscErrorCode MatGetState(Mat A, PetscObjectState *state)
894: {
895: PetscFunctionBegin;
897: PetscAssertPointer(state, 2);
898: PetscCall(PetscObjectStateGet((PetscObject)A, state));
899: PetscFunctionReturn(PETSC_SUCCESS);
900: }
902: /*@
903: MatResetPreallocation - Reset matrix to use the original preallocation values provided by the user, for example with `MatXAIJSetPreallocation()`
905: Collective
907: Input Parameter:
908: . A - the matrix
910: Level: beginner
912: Notes:
913: After calling `MatAssemblyBegin()` and `MatAssemblyEnd()` with `MAT_FINAL_ASSEMBLY` the matrix data structures represent the nonzeros assigned to the
914: matrix. If that space is less than the preallocated space that extra preallocated space is no longer available to take on new values. `MatResetPreallocation()`
915: makes all of the preallocation space available
917: Current values in the matrix are lost in this call.
919: Currently only supported for `MATAIJ` matrices.
921: .seealso: [](ch_matrices), `Mat`, `MatSeqAIJSetPreallocation()`, `MatMPIAIJSetPreallocation()`, `MatXAIJSetPreallocation()`
922: @*/
923: PetscErrorCode MatResetPreallocation(Mat A)
924: {
925: PetscFunctionBegin;
928: PetscCheck(A->insertmode == NOT_SET_VALUES, PETSC_COMM_SELF, PETSC_ERR_SUP, "Cannot reset preallocation after setting some values but not yet calling MatAssemblyBegin()/MatAssemblyEnd()");
929: if (A->num_ass == 0) PetscFunctionReturn(PETSC_SUCCESS);
930: PetscUseMethod(A, "MatResetPreallocation_C", (Mat), (A));
931: PetscFunctionReturn(PETSC_SUCCESS);
932: }
934: /*@
935: MatSetUp - Sets up the internal matrix data structures for later use by the matrix
937: Collective
939: Input Parameter:
940: . A - the matrix
942: Level: advanced
944: Notes:
945: If the user has not set preallocation for this matrix then an efficient algorithm will be used for the first round of
946: setting values in the matrix.
948: This routine is called internally by other `Mat` functions when needed so rarely needs to be called by users
950: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatCreate()`, `MatDestroy()`, `MatXAIJSetPreallocation()`
951: @*/
952: PetscErrorCode MatSetUp(Mat A)
953: {
954: PetscFunctionBegin;
956: if (!((PetscObject)A)->type_name) {
957: PetscMPIInt size;
959: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A), &size));
960: PetscCall(MatSetType(A, size == 1 ? MATSEQAIJ : MATMPIAIJ));
961: }
962: if (!A->preallocated) PetscTryTypeMethod(A, setup);
963: PetscCall(PetscLayoutSetUp(A->rmap));
964: PetscCall(PetscLayoutSetUp(A->cmap));
965: A->preallocated = PETSC_TRUE;
966: PetscFunctionReturn(PETSC_SUCCESS);
967: }
969: #if defined(PETSC_HAVE_SAWS)
970: #include <petscviewersaws.h>
971: #endif
973: /*
974: If threadsafety is on extraneous matrices may be printed
976: This flag cannot be stored in the matrix because the original matrix in MatView() may assemble a new matrix which is passed into MatViewFromOptions()
977: */
978: #if !defined(PETSC_HAVE_THREADSAFETY)
979: static PetscInt insidematview = 0;
980: #endif
982: /*@
983: MatViewFromOptions - View properties of the matrix based on options set in the options database
985: Collective
987: Input Parameters:
988: + A - the matrix
989: . obj - optional additional object that provides the options prefix to use
990: - name - command line option
992: Options Database Key:
993: . -mat_view [viewertype]:... - the viewer and its options
995: Level: intermediate
997: Note:
998: .vb
999: If no value is provided ascii:stdout is used
1000: ascii[:[filename][:[format][:append]]] defaults to stdout - format can be one of ascii_info, ascii_info_detail, or ascii_matlab,
1001: for example ascii::ascii_info prints just the information about the object not all details
1002: unless :append is given filename opens in write mode, overwriting what was already there
1003: binary[:[filename][:[format][:append]]] defaults to the file binaryoutput
1004: draw[:drawtype[:filename]] for example, draw:tikz, draw:tikz:figure.tex or draw:x
1005: socket[:port] defaults to the standard output port
1006: saws[:communicatorname] publishes object to the Scientific Application Webserver (SAWs)
1007: .ve
1009: .seealso: [](ch_matrices), `Mat`, `MatView()`, `PetscObjectViewFromOptions()`, `MatCreate()`
1010: @*/
1011: PetscErrorCode MatViewFromOptions(Mat A, PetscObject obj, const char name[])
1012: {
1013: PetscFunctionBegin;
1015: #if !defined(PETSC_HAVE_THREADSAFETY)
1016: if (insidematview) PetscFunctionReturn(PETSC_SUCCESS);
1017: #endif
1018: PetscCall(PetscObjectViewFromOptions((PetscObject)A, obj, name));
1019: PetscFunctionReturn(PETSC_SUCCESS);
1020: }
1022: /*@
1023: MatView - display information about a matrix in a variety ways
1025: Collective on viewer
1027: Input Parameters:
1028: + mat - the matrix
1029: - viewer - visualization context
1031: Options Database Keys:
1032: + -mat_view ::ascii_info - Prints info on matrix at conclusion of `MatAssemblyEnd()`
1033: . -mat_view ::ascii_info_detail - Prints more detailed info
1034: . -mat_view - Prints matrix in ASCII format
1035: . -mat_view ::ascii_matlab - Prints matrix in MATLAB format
1036: . -mat_view draw - PetscDraws nonzero structure of matrix, using `MatView()` and `PetscDrawOpenX()`.
1037: . -display <name> - Sets display name (default is host)
1038: . -draw_pause <sec> - Sets number of seconds to pause after display
1039: . -mat_view socket - Sends matrix to socket, can be accessed from MATLAB (see Users-Manual: ch_matlab for details)
1040: . -viewer_socket_machine <machine> - -
1041: . -viewer_socket_port <port> - -
1042: . -mat_view binary - save matrix to file in binary format
1043: - -viewer_binary_filename <name> - -
1045: Level: beginner
1047: Notes:
1048: The available visualization contexts include
1049: + `PETSC_VIEWER_STDOUT_SELF` - for sequential matrices
1050: . `PETSC_VIEWER_STDOUT_WORLD` - for parallel matrices created on `PETSC_COMM_WORLD`
1051: . `PETSC_VIEWER_STDOUT_`(comm) - for matrices created on MPI communicator comm
1052: - `PETSC_VIEWER_DRAW_WORLD` - graphical display of nonzero structure
1054: The user can open alternative visualization contexts with
1055: + `PetscViewerASCIIOpen()` - Outputs matrix to a specified file
1056: . `PetscViewerBinaryOpen()` - Outputs matrix in binary to a specified file; corresponding input uses `MatLoad()`
1057: . `PetscViewerDrawOpen()` - Outputs nonzero matrix nonzero structure to an X window display
1058: - `PetscViewerSocketOpen()` - Outputs matrix to Socket viewer, `PETSCVIEWERSOCKET`. Only the `MATSEQDENSE` and `MATAIJ` types support this viewer.
1060: The user can call `PetscViewerPushFormat()` to specify the output
1061: format of ASCII printed objects (when using `PETSC_VIEWER_STDOUT_SELF`,
1062: `PETSC_VIEWER_STDOUT_WORLD` and `PetscViewerASCIIOpen()`). Available formats include
1063: + `PETSC_VIEWER_DEFAULT` - default, prints matrix contents
1064: . `PETSC_VIEWER_ASCII_MATLAB` - prints matrix contents in MATLAB format
1065: . `PETSC_VIEWER_ASCII_DENSE` - prints entire matrix including zeros
1066: . `PETSC_VIEWER_ASCII_COMMON` - prints matrix contents, using a sparse format common among all matrix types
1067: . `PETSC_VIEWER_ASCII_IMPL` - prints matrix contents, using an implementation-specific format (which is in many cases the same as the default)
1068: . `PETSC_VIEWER_ASCII_INFO` - prints basic information about the matrix size and structure (not the matrix entries)
1069: - `PETSC_VIEWER_ASCII_INFO_DETAIL` - prints more detailed information about the matrix nonzero structure (still not vector or matrix entries)
1071: The ASCII viewers are only recommended for small matrices on at most a moderate number of processes,
1072: the program will seemingly hang and take hours for larger matrices, for larger matrices one should use the binary format.
1074: In the debugger you can do "call MatView(mat,0)" to display the matrix. (The same holds for any PETSc object viewer).
1076: See the manual page for `MatLoad()` for the exact format of the binary file when the binary
1077: viewer is used.
1079: See share/petsc/matlab/PetscBinaryRead.m for a MATLAB code that can read in the binary file when the binary
1080: viewer is used and lib/petsc/bin/PetscBinaryIO.py for loading them into Python.
1082: One can use '-mat_view draw -draw_pause -1' to pause the graphical display of matrix nonzero structure,
1083: and then use the following mouse functions.
1084: .vb
1085: left mouse: zoom in
1086: middle mouse: zoom out
1087: right mouse: continue with the simulation
1088: .ve
1090: .seealso: [](ch_matrices), `Mat`, `PetscViewerPushFormat()`, `PetscViewerASCIIOpen()`, `PetscViewerDrawOpen()`, `PetscViewer`,
1091: `PetscViewerSocketOpen()`, `PetscViewerBinaryOpen()`, `MatLoad()`, `MatViewFromOptions()`
1092: @*/
1093: PetscErrorCode MatView(Mat mat, PetscViewer viewer)
1094: {
1095: PetscInt rows, cols, rbs, cbs;
1096: PetscBool isascii, isstring, issaws;
1097: PetscViewerFormat format;
1098: PetscMPIInt size;
1100: PetscFunctionBegin;
1103: if (!viewer) PetscCall(PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)mat), &viewer));
1106: PetscCall(PetscViewerGetFormat(viewer, &format));
1107: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)viewer), &size));
1108: if (size == 1 && format == PETSC_VIEWER_LOAD_BALANCE) PetscFunctionReturn(PETSC_SUCCESS);
1110: #if !defined(PETSC_HAVE_THREADSAFETY)
1111: insidematview++;
1112: #endif
1113: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSTRING, &isstring));
1114: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
1115: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSAWS, &issaws));
1116: 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");
1118: PetscCall(PetscLogEventBegin(MAT_View, mat, viewer, 0, 0));
1119: if (isascii) {
1120: if (!mat->preallocated) {
1121: PetscCall(PetscViewerASCIIPrintf(viewer, "Matrix has not been preallocated yet\n"));
1122: #if !defined(PETSC_HAVE_THREADSAFETY)
1123: insidematview--;
1124: #endif
1125: PetscCall(PetscLogEventEnd(MAT_View, mat, viewer, 0, 0));
1126: PetscFunctionReturn(PETSC_SUCCESS);
1127: }
1128: if (!mat->assembled) {
1129: PetscCall(PetscViewerASCIIPrintf(viewer, "Matrix has not been assembled yet\n"));
1130: #if !defined(PETSC_HAVE_THREADSAFETY)
1131: insidematview--;
1132: #endif
1133: PetscCall(PetscLogEventEnd(MAT_View, mat, viewer, 0, 0));
1134: PetscFunctionReturn(PETSC_SUCCESS);
1135: }
1136: PetscCall(PetscObjectPrintClassNamePrefixType((PetscObject)mat, viewer));
1137: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1138: MatNullSpace nullsp, transnullsp;
1140: PetscCall(PetscViewerASCIIPushTab(viewer));
1141: PetscCall(MatGetSize(mat, &rows, &cols));
1142: PetscCall(MatGetBlockSizes(mat, &rbs, &cbs));
1143: if (rbs != 1 || cbs != 1) {
1144: 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" : ""));
1145: else PetscCall(PetscViewerASCIIPrintf(viewer, "rows=%" PetscInt_FMT ", cols=%" PetscInt_FMT ", bs=%" PetscInt_FMT "%s\n", rows, cols, rbs, mat->bsizes ? " variable blocks set" : ""));
1146: } else PetscCall(PetscViewerASCIIPrintf(viewer, "rows=%" PetscInt_FMT ", cols=%" PetscInt_FMT "\n", rows, cols));
1147: if (mat->factortype) {
1148: MatSolverType solver;
1149: PetscCall(MatFactorGetSolverType(mat, &solver));
1150: PetscCall(PetscViewerASCIIPrintf(viewer, "package used to perform factorization: %s\n", solver));
1151: }
1152: if (mat->ops->getinfo) {
1153: MatInfo info;
1154: PetscCall(MatGetInfo(mat, MAT_GLOBAL_SUM, &info));
1155: PetscCall(PetscViewerASCIIPrintf(viewer, "total: nonzeros=%.f, allocated nonzeros=%.f\n", info.nz_used, info.nz_allocated));
1156: if (!mat->factortype) PetscCall(PetscViewerASCIIPrintf(viewer, "total number of mallocs used during MatSetValues calls=%" PetscInt_FMT "\n", (PetscInt)info.mallocs));
1157: }
1158: PetscCall(MatGetNullSpace(mat, &nullsp));
1159: PetscCall(MatGetTransposeNullSpace(mat, &transnullsp));
1160: if (nullsp) PetscCall(PetscViewerASCIIPrintf(viewer, " has attached null space\n"));
1161: if (transnullsp && transnullsp != nullsp) PetscCall(PetscViewerASCIIPrintf(viewer, " has attached transposed null space\n"));
1162: PetscCall(MatGetNearNullSpace(mat, &nullsp));
1163: if (nullsp) PetscCall(PetscViewerASCIIPrintf(viewer, " has attached near null space\n"));
1164: PetscCall(PetscViewerASCIIPushTab(viewer));
1165: PetscCall(MatProductView(mat, viewer));
1166: PetscCall(PetscViewerASCIIPopTab(viewer));
1167: if (mat->bsizes && format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1168: IS tmp;
1170: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)viewer), mat->nblocks, mat->bsizes, PETSC_USE_POINTER, &tmp));
1171: PetscCall(PetscObjectSetName((PetscObject)tmp, "Block Sizes"));
1172: PetscCall(PetscViewerASCIIPushTab(viewer));
1173: PetscCall(ISView(tmp, viewer));
1174: PetscCall(PetscViewerASCIIPopTab(viewer));
1175: PetscCall(ISDestroy(&tmp));
1176: }
1177: }
1178: } else if (issaws) {
1179: #if defined(PETSC_HAVE_SAWS)
1180: PetscMPIInt rank;
1182: PetscCall(PetscObjectName((PetscObject)mat));
1183: PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
1184: if (!((PetscObject)mat)->amsmem && rank == 0) PetscCall(PetscObjectViewSAWs((PetscObject)mat, viewer));
1185: #endif
1186: } else if (isstring) {
1187: const char *type;
1188: PetscCall(MatGetType(mat, &type));
1189: PetscCall(PetscViewerStringSPrintf(viewer, " MatType: %-7.7s", type));
1190: PetscTryTypeMethod(mat, view, viewer);
1191: }
1192: if ((format == PETSC_VIEWER_NATIVE || format == PETSC_VIEWER_LOAD_BALANCE) && mat->ops->viewnative) {
1193: PetscCall(PetscViewerASCIIPushTab(viewer));
1194: PetscUseTypeMethod(mat, viewnative, viewer);
1195: PetscCall(PetscViewerASCIIPopTab(viewer));
1196: } else if (mat->ops->view) {
1197: PetscCall(PetscViewerASCIIPushTab(viewer));
1198: PetscUseTypeMethod(mat, view, viewer);
1199: PetscCall(PetscViewerASCIIPopTab(viewer));
1200: }
1201: if (isascii) {
1202: PetscCall(PetscViewerGetFormat(viewer, &format));
1203: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) PetscCall(PetscViewerASCIIPopTab(viewer));
1204: }
1205: PetscCall(PetscLogEventEnd(MAT_View, mat, viewer, 0, 0));
1206: #if !defined(PETSC_HAVE_THREADSAFETY)
1207: insidematview--;
1208: #endif
1209: PetscFunctionReturn(PETSC_SUCCESS);
1210: }
1212: #if defined(PETSC_USE_DEBUG)
1213: #include <../src/sys/totalview/tv_data_display.h>
1214: PETSC_UNUSED static int TV_display_type(const struct _p_Mat *mat)
1215: {
1216: TV_add_row("Local rows", "int", &mat->rmap->n);
1217: TV_add_row("Local columns", "int", &mat->cmap->n);
1218: TV_add_row("Global rows", "int", &mat->rmap->N);
1219: TV_add_row("Global columns", "int", &mat->cmap->N);
1220: TV_add_row("Typename", TV_ascii_string_type, ((PetscObject)mat)->type_name);
1221: return TV_format_OK;
1222: }
1223: #endif
1225: /*@
1226: MatLoad - Loads a matrix that has been stored in binary/HDF5 format
1227: with `MatView()`. The matrix format is determined from the options database.
1228: Generates a parallel MPI matrix if the communicator has more than one
1229: processor. The default matrix type is `MATAIJ`.
1231: Collective
1233: Input Parameters:
1234: + mat - the newly loaded matrix, this needs to have been created with `MatCreate()`
1235: or some related function before a call to `MatLoad()`
1236: - viewer - `PETSCVIEWERBINARY`/`PETSCVIEWERHDF5` file viewer
1238: Options Database Key:
1239: . -matload_block_size <bs> - set block size
1241: Level: beginner
1243: Notes:
1244: If the `Mat` type has not yet been given then `MATAIJ` is used, call `MatSetFromOptions()` on the
1245: `Mat` before calling this routine if you wish to set it from the options database.
1247: `MatLoad()` automatically loads into the options database any options
1248: given in the file filename.info where filename is the name of the file
1249: that was passed to the `PetscViewerBinaryOpen()`. The options in the info
1250: file will be ignored if you use the -viewer_binary_skip_info option.
1252: If the type or size of mat is not set before a call to `MatLoad()`, PETSc
1253: sets the default matrix type AIJ and sets the local and global sizes.
1254: If type and/or size is already set, then the same are used.
1256: In parallel, each processor can load a subset of rows (or the
1257: entire matrix). This routine is especially useful when a large
1258: matrix is stored on disk and only part of it is desired on each
1259: processor. For example, a parallel solver may access only some of
1260: the rows from each processor. The algorithm used here reads
1261: relatively small blocks of data rather than reading the entire
1262: matrix and then subsetting it.
1264: Viewer's `PetscViewerType` must be either `PETSCVIEWERBINARY` or `PETSCVIEWERHDF5`.
1265: Such viewer can be created using `PetscViewerBinaryOpen()` or `PetscViewerHDF5Open()`,
1266: or the sequence like
1267: .vb
1268: `PetscViewer` v;
1269: `PetscViewerCreate`(`PETSC_COMM_WORLD`,&v);
1270: `PetscViewerSetType`(v,`PETSCVIEWERBINARY`);
1271: `PetscViewerSetFromOptions`(v);
1272: `PetscViewerFileSetMode`(v,`FILE_MODE_READ`);
1273: `PetscViewerFileSetName`(v,"datafile");
1274: .ve
1275: The optional `PetscViewerSetFromOptions()` call allows overriding `PetscViewerSetType()` using the option
1276: $ -viewer_type {binary, hdf5}
1278: See the example src/ksp/ksp/tutorials/ex27.c with the first approach,
1279: and src/mat/tutorials/ex10.c with the second approach.
1281: In case of `PETSCVIEWERBINARY`, a native PETSc binary format is used. Each of the blocks
1282: is read onto MPI rank 0 and then shipped to its destination MPI rank, one after another.
1283: Multiple objects, both matrices and vectors, can be stored within the same file.
1284: Their `PetscObject` name is ignored; they are loaded in the order of their storage.
1286: Most users should not need to know the details of the binary storage
1287: format, since `MatLoad()` and `MatView()` completely hide these details.
1288: But for anyone who is interested, the standard binary matrix storage
1289: format is
1291: .vb
1292: PetscInt MAT_FILE_CLASSID
1293: PetscInt number of rows
1294: PetscInt number of columns
1295: PetscInt total number of nonzeros
1296: PetscInt *number nonzeros in each row
1297: PetscInt *column indices of all nonzeros (starting index is zero)
1298: PetscScalar *values of all nonzeros
1299: .ve
1300: If PETSc was not configured with `--with-64-bit-indices` then only `MATMPIAIJ` matrices with more than `PETSC_INT_MAX` non-zeros can be
1301: 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
1302: case will not fit in a (32-bit) `PetscInt` the value `PETSC_INT_MAX` is used for the header entry `total number of nonzeros`.
1304: PETSc automatically does the byte swapping for
1305: machines that store the bytes reversed. Thus if you write your own binary
1306: read/write routines you have to swap the bytes; see `PetscBinaryRead()`
1307: and `PetscBinaryWrite()` to see how this may be done.
1309: In case of `PETSCVIEWERHDF5`, a parallel HDF5 reader is used.
1310: Each processor's chunk is loaded independently by its owning MPI process.
1311: Multiple objects, both matrices and vectors, can be stored within the same file.
1312: They are looked up by their PetscObject name.
1314: As the MATLAB MAT-File Version 7.3 format is also a HDF5 flavor, we decided to use
1315: by default the same structure and naming of the AIJ arrays and column count
1316: within the HDF5 file. This means that a MAT file saved with -v7.3 flag, e.g.
1317: $ save example.mat A b -v7.3
1318: can be directly read by this routine (see Reference 1 for details).
1320: Depending on your MATLAB version, this format might be a default,
1321: otherwise you can set it as default in Preferences.
1323: Unless -nocompression flag is used to save the file in MATLAB,
1324: PETSc must be configured with ZLIB package.
1326: See also examples src/mat/tutorials/ex10.c and src/ksp/ksp/tutorials/ex27.c
1328: This reader currently supports only real `MATSEQAIJ`, `MATMPIAIJ`, `MATSEQDENSE` and `MATMPIDENSE` matrices for `PETSCVIEWERHDF5`
1330: Corresponding `MatView()` is not yet implemented.
1332: The loaded matrix is actually a transpose of the original one in MATLAB,
1333: unless you push `PETSC_VIEWER_HDF5_MAT` format (see examples above).
1334: With this format, matrix is automatically transposed by PETSc,
1335: unless the matrix is marked as SPD or symmetric
1336: (see `MatSetOption()`, `MAT_SPD`, `MAT_SYMMETRIC`).
1338: See MATLAB Documentation on `save()`, <https://www.mathworks.com/help/matlab/ref/save.html#btox10b-1-version>
1340: .seealso: [](ch_matrices), `Mat`, `PetscViewerBinaryOpen()`, `PetscViewerSetType()`, `MatView()`, `VecLoad()`
1341: @*/
1342: PetscErrorCode MatLoad(Mat mat, PetscViewer viewer)
1343: {
1344: PetscBool flg;
1346: PetscFunctionBegin;
1350: if (!((PetscObject)mat)->type_name) PetscCall(MatSetType(mat, MATAIJ));
1352: flg = PETSC_FALSE;
1353: PetscCall(PetscOptionsGetBool(((PetscObject)mat)->options, ((PetscObject)mat)->prefix, "-matload_symmetric", &flg, NULL));
1354: if (flg) {
1355: PetscCall(MatSetOption(mat, MAT_SYMMETRIC, PETSC_TRUE));
1356: PetscCall(MatSetOption(mat, MAT_SYMMETRY_ETERNAL, PETSC_TRUE));
1357: }
1358: flg = PETSC_FALSE;
1359: PetscCall(PetscOptionsGetBool(((PetscObject)mat)->options, ((PetscObject)mat)->prefix, "-matload_spd", &flg, NULL));
1360: if (flg) PetscCall(MatSetOption(mat, MAT_SPD, PETSC_TRUE));
1362: PetscCall(PetscLogEventBegin(MAT_Load, mat, viewer, 0, 0));
1363: PetscUseTypeMethod(mat, load, viewer);
1364: PetscCall(PetscLogEventEnd(MAT_Load, mat, viewer, 0, 0));
1365: PetscFunctionReturn(PETSC_SUCCESS);
1366: }
1368: static PetscErrorCode MatDestroy_Redundant(Mat_Redundant **redundant)
1369: {
1370: Mat_Redundant *redund = *redundant;
1372: PetscFunctionBegin;
1373: if (redund) {
1374: if (redund->matseq) { /* via MatCreateSubMatrices() */
1375: PetscCall(ISDestroy(&redund->isrow));
1376: PetscCall(ISDestroy(&redund->iscol));
1377: PetscCall(MatDestroySubMatrices(1, &redund->matseq));
1378: } else {
1379: PetscCall(PetscFree2(redund->send_rank, redund->recv_rank));
1380: PetscCall(PetscFree(redund->sbuf_j));
1381: PetscCall(PetscFree(redund->sbuf_a));
1382: for (PetscInt i = 0; i < redund->nrecvs; i++) {
1383: PetscCall(PetscFree(redund->rbuf_j[i]));
1384: PetscCall(PetscFree(redund->rbuf_a[i]));
1385: }
1386: PetscCall(PetscFree4(redund->sbuf_nz, redund->rbuf_nz, redund->rbuf_j, redund->rbuf_a));
1387: }
1389: if (redund->subcomm) PetscCall(PetscCommDestroy(&redund->subcomm));
1390: PetscCall(PetscFree(redund));
1391: }
1392: PetscFunctionReturn(PETSC_SUCCESS);
1393: }
1395: /*@
1396: MatDestroy - Frees space taken by a matrix.
1398: Collective
1400: Input Parameter:
1401: . A - the matrix
1403: Level: beginner
1405: Developer Note:
1406: Some special arrays of matrices are not destroyed in this routine but instead by the routines called by
1407: `MatDestroySubMatrices()`. Thus one must be sure that any changes here must also be made in those routines.
1408: `MatHeaderMerge()` and `MatHeaderReplace()` also manipulate the data in the `Mat` object and likely need changes
1409: if changes are needed here.
1411: .seealso: [](ch_matrices), `Mat`, `MatCreate()`
1412: @*/
1413: PetscErrorCode MatDestroy(Mat *A)
1414: {
1415: PetscFunctionBegin;
1416: if (!*A) PetscFunctionReturn(PETSC_SUCCESS);
1418: if (--((PetscObject)*A)->refct > 0) {
1419: *A = NULL;
1420: PetscFunctionReturn(PETSC_SUCCESS);
1421: }
1423: /* if memory was published with SAWs then destroy it */
1424: PetscCall(PetscObjectSAWsViewOff((PetscObject)*A));
1425: PetscTryTypeMethod(*A, destroy);
1427: PetscCall(PetscFree((*A)->factorprefix));
1428: PetscCall(PetscFree((*A)->defaultvectype));
1429: PetscCall(PetscFree((*A)->defaultrandtype));
1430: PetscCall(PetscFree((*A)->bsizes));
1431: PetscCall(PetscFree((*A)->solvertype));
1432: for (PetscInt i = 0; i < MAT_FACTOR_NUM_TYPES; i++) PetscCall(PetscFree((*A)->preferredordering[i]));
1433: if ((*A)->redundant && (*A)->redundant->matseq[0] == *A) (*A)->redundant->matseq[0] = NULL;
1434: PetscCall(MatDestroy_Redundant(&(*A)->redundant));
1435: PetscCall(MatProductClear(*A));
1436: PetscCall(MatNullSpaceDestroy(&(*A)->nullsp));
1437: PetscCall(MatNullSpaceDestroy(&(*A)->transnullsp));
1438: PetscCall(MatNullSpaceDestroy(&(*A)->nearnullsp));
1439: PetscCall(MatDestroy(&(*A)->schur));
1440: PetscCall(PetscLayoutDestroy(&(*A)->rmap));
1441: PetscCall(PetscLayoutDestroy(&(*A)->cmap));
1442: PetscCall(PetscHeaderDestroy(A));
1443: PetscFunctionReturn(PETSC_SUCCESS);
1444: }
1446: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
1447: /*@
1448: MatSetValues - Inserts or adds a block of values into a matrix.
1449: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
1450: MUST be called after all calls to `MatSetValues()` have been completed.
1452: Not Collective
1454: Input Parameters:
1455: + mat - the matrix
1456: . v - a logically two-dimensional array of values
1457: . m - the number of rows
1458: . idxm - the global indices of the rows
1459: . n - the number of columns
1460: . idxn - the global indices of the columns
1461: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
1463: Level: beginner
1465: Notes:
1466: By default the values, `v`, are stored row-oriented. See `MatSetOption()` for other options.
1468: Calls to `MatSetValues()` with the `INSERT_VALUES` and `ADD_VALUES`
1469: options cannot be mixed without intervening calls to the assembly
1470: routines.
1472: `MatSetValues()` uses 0-based row and column numbers in Fortran
1473: as well as in C.
1475: Negative indices may be passed in `idxm` and `idxn`, these rows and columns are
1476: simply ignored. This allows easily inserting element stiffness matrices
1477: with homogeneous Dirichlet boundary conditions that you don't want represented
1478: in the matrix.
1480: Efficiency Alert:
1481: The routine `MatSetValuesBlocked()` may offer much better efficiency
1482: for users of block sparse formats (`MATSEQBAIJ` and `MATMPIBAIJ`).
1484: Fortran Notes:
1485: If any of `idxm`, `idxn`, and `v` are scalars pass them using, for example,
1486: .vb
1487: MatSetValues(mat, one, [idxm], one, [idxn], [v], INSERT_VALUES)
1488: .ve
1490: If `v` is a two-dimensional array use `reshape()` to pass it as a one dimensional array
1492: Developer Note:
1493: This is labeled with C so does not automatically generate Fortran stubs and interfaces
1494: because it requires multiple Fortran interfaces depending on which arguments are scalar or arrays.
1496: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1497: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`
1498: @*/
1499: PetscErrorCode MatSetValues(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], const PetscScalar v[], InsertMode addv)
1500: {
1501: PetscFunctionBeginHot;
1504: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
1505: PetscAssertPointer(idxm, 3);
1506: PetscAssertPointer(idxn, 5);
1507: MatCheckPreallocated(mat, 1);
1509: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
1510: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
1512: if (PetscDefined(USE_DEBUG)) {
1513: PetscInt i, j;
1515: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
1516: if (v) {
1517: for (i = 0; i < m; i++) {
1518: for (j = 0; j < n; j++) {
1519: if (mat->erroriffailure && PetscIsInfOrNanScalar(v[i * n + j]))
1520: #if defined(PETSC_USE_COMPLEX)
1521: 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]);
1522: #else
1523: 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]);
1524: #endif
1525: }
1526: }
1527: }
1528: 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);
1529: 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);
1530: }
1532: if (mat->assembled) {
1533: mat->was_assembled = PETSC_TRUE;
1534: mat->assembled = PETSC_FALSE;
1535: }
1536: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
1537: PetscUseTypeMethod(mat, setvalues, m, idxm, n, idxn, v, addv);
1538: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
1539: PetscFunctionReturn(PETSC_SUCCESS);
1540: }
1542: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
1543: /*@
1544: MatSetValuesIS - Inserts or adds a block of values into a matrix using an `IS` to indicate the rows and columns
1545: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
1546: MUST be called after all calls to `MatSetValues()` have been completed.
1548: Not Collective
1550: Input Parameters:
1551: + mat - the matrix
1552: . v - a logically two-dimensional array of values
1553: . ism - the rows to provide
1554: . isn - the columns to provide
1555: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
1557: Level: beginner
1559: Notes:
1560: By default the values, `v`, are stored row-oriented. See `MatSetOption()` for other options.
1562: Calls to `MatSetValues()` with the `INSERT_VALUES` and `ADD_VALUES`
1563: options cannot be mixed without intervening calls to the assembly
1564: routines.
1566: `MatSetValues()` uses 0-based row and column numbers in Fortran
1567: as well as in C.
1569: Negative indices may be passed in `ism` and `isn`, these rows and columns are
1570: simply ignored. This allows easily inserting element stiffness matrices
1571: with homogeneous Dirichlet boundary conditions that you don't want represented
1572: in the matrix.
1574: Efficiency Alert:
1575: The routine `MatSetValuesBlocked()` may offer much better efficiency
1576: for users of block sparse formats (`MATSEQBAIJ` and `MATMPIBAIJ`).
1578: This is currently not optimized for any particular `ISType`
1580: Developer Note:
1581: This is labeled with C so does not automatically generate Fortran stubs and interfaces
1582: because it requires multiple Fortran interfaces depending on which arguments are scalar or arrays.
1584: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatSetValues()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1585: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`
1586: @*/
1587: PetscErrorCode MatSetValuesIS(Mat mat, IS ism, IS isn, const PetscScalar v[], InsertMode addv)
1588: {
1589: PetscInt m, n;
1590: const PetscInt *rows, *cols;
1592: PetscFunctionBeginHot;
1594: PetscCall(ISGetIndices(ism, &rows));
1595: PetscCall(ISGetIndices(isn, &cols));
1596: PetscCall(ISGetLocalSize(ism, &m));
1597: PetscCall(ISGetLocalSize(isn, &n));
1598: PetscCall(MatSetValues(mat, m, rows, n, cols, v, addv));
1599: PetscCall(ISRestoreIndices(ism, &rows));
1600: PetscCall(ISRestoreIndices(isn, &cols));
1601: PetscFunctionReturn(PETSC_SUCCESS);
1602: }
1604: /*@
1605: MatSetValuesRowLocal - Inserts a row (block row for `MATBAIJ` matrices) of nonzero
1606: values into a matrix
1608: Not Collective
1610: Input Parameters:
1611: + mat - the matrix
1612: . row - the (block) row to set
1613: - v - a logically two-dimensional array of values
1615: Level: intermediate
1617: Notes:
1618: The values, `v`, are column-oriented (for the block version) and sorted
1620: All the nonzero values in `row` must be provided
1622: The matrix must have previously had its column indices set, likely by having been assembled.
1624: `row` must belong to this MPI process
1626: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1627: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`, `MatSetValues()`, `MatSetValuesRow()`, `MatSetLocalToGlobalMapping()`
1628: @*/
1629: PetscErrorCode MatSetValuesRowLocal(Mat mat, PetscInt row, const PetscScalar v[])
1630: {
1631: PetscInt globalrow;
1633: PetscFunctionBegin;
1636: PetscAssertPointer(v, 3);
1637: PetscCall(ISLocalToGlobalMappingApply(mat->rmap->mapping, 1, &row, &globalrow));
1638: PetscCall(MatSetValuesRow(mat, globalrow, v));
1639: PetscFunctionReturn(PETSC_SUCCESS);
1640: }
1642: /*@
1643: MatSetValuesRow - Inserts a row (block row for `MATBAIJ` matrices) of nonzero
1644: values into a matrix
1646: Not Collective
1648: Input Parameters:
1649: + mat - the matrix
1650: . row - the (block) row to set
1651: - v - a logically two-dimensional (column major) array of values for block matrices with blocksize larger than one, otherwise a one dimensional array of values
1653: Level: advanced
1655: Notes:
1656: The values, `v`, are column-oriented for the block version.
1658: All the nonzeros in `row` must be provided
1660: THE MATRIX MUST HAVE PREVIOUSLY HAD ITS COLUMN INDICES SET. IT IS RARE THAT THIS ROUTINE IS USED, usually `MatSetValues()` is used.
1662: `row` must belong to this process
1664: .seealso: [](ch_matrices), `Mat`, `MatSetValues()`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1665: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`
1666: @*/
1667: PetscErrorCode MatSetValuesRow(Mat mat, PetscInt row, const PetscScalar v[])
1668: {
1669: PetscFunctionBeginHot;
1672: MatCheckPreallocated(mat, 1);
1673: PetscAssertPointer(v, 3);
1674: PetscCheck(mat->insertmode != ADD_VALUES, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add and insert values");
1675: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
1676: mat->insertmode = INSERT_VALUES;
1678: if (mat->assembled) {
1679: mat->was_assembled = PETSC_TRUE;
1680: mat->assembled = PETSC_FALSE;
1681: }
1682: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
1683: PetscUseTypeMethod(mat, setvaluesrow, row, v);
1684: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
1685: PetscFunctionReturn(PETSC_SUCCESS);
1686: }
1688: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
1689: /*@
1690: MatSetValuesStencil - Inserts or adds a block of values into a matrix.
1691: Using structured grid indexing
1693: Not Collective
1695: Input Parameters:
1696: + mat - the matrix
1697: . m - number of rows being entered
1698: . idxm - grid coordinates (and component number when dof > 1) for matrix rows being entered
1699: . n - number of columns being entered
1700: . idxn - grid coordinates (and component number when dof > 1) for matrix columns being entered
1701: . v - a logically two-dimensional array of values
1702: - addv - either `ADD_VALUES` to add to existing entries at that location or `INSERT_VALUES` to replace existing entries with new values
1704: Level: beginner
1706: Notes:
1707: By default the values, `v`, are row-oriented. See `MatSetOption()` for other options.
1709: Calls to `MatSetValuesStencil()` with the `INSERT_VALUES` and `ADD_VALUES`
1710: options cannot be mixed without intervening calls to the assembly
1711: routines.
1713: The grid coordinates are across the entire grid, not just the local portion
1715: `MatSetValuesStencil()` uses 0-based row and column numbers in Fortran
1716: as well as in C.
1718: For setting/accessing vector values via array coordinates you can use the `DMDAVecGetArray()` routine
1720: In order to use this routine you must either obtain the matrix with `DMCreateMatrix()`
1721: or call `MatSetLocalToGlobalMapping()` and `MatSetStencil()` first.
1723: The columns and rows in the stencil passed in MUST be contained within the
1724: ghost region of the given process as set with DMDACreateXXX() or `MatSetStencil()`. For example,
1725: if you create a `DMDA` with an overlap of one grid level and on a particular process its first
1726: local nonghost x logical coordinate is 6 (so its first ghost x logical coordinate is 5) the
1727: first i index you can use in your column and row indices in `MatSetStencil()` is 5.
1729: For periodic boundary conditions use negative indices for values to the left (below 0; that are to be
1730: obtained by wrapping values from right edge). For values to the right of the last entry using that index plus one
1731: etc to obtain values that obtained by wrapping the values from the left edge. This does not work for anything but the
1732: `DM_BOUNDARY_PERIODIC` boundary type.
1734: 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
1735: a single value per point) you can skip filling those indices.
1737: Inspired by the structured grid interface to the HYPRE package
1738: (https://computation.llnl.gov/projects/hypre-scalable-linear-solvers-multigrid-methods)
1740: Efficiency Alert:
1741: The routine `MatSetValuesBlockedStencil()` may offer much better efficiency
1742: for users of block sparse formats (`MATSEQBAIJ` and `MATMPIBAIJ`).
1744: Fortran Note:
1745: `idxm` and `idxn` should be declared as
1746: $ MatStencil idxm(4,m),idxn(4,n)
1747: and the values inserted using
1748: .vb
1749: idxm(MatStencil_i,1) = i
1750: idxm(MatStencil_j,1) = j
1751: idxm(MatStencil_k,1) = k
1752: idxm(MatStencil_c,1) = c
1753: etc
1754: .ve
1756: .seealso: [](ch_matrices), `Mat`, `DMDA`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`
1757: `MatSetValues()`, `MatSetValuesBlockedStencil()`, `MatSetStencil()`, `DMCreateMatrix()`, `DMDAVecGetArray()`, `MatStencil`
1758: @*/
1759: PetscErrorCode MatSetValuesStencil(Mat mat, PetscInt m, const MatStencil idxm[], PetscInt n, const MatStencil idxn[], const PetscScalar v[], InsertMode addv)
1760: {
1761: PetscInt buf[8192], *bufm = NULL, *bufn = NULL, *jdxm, *jdxn;
1762: PetscInt j, i, dim = mat->stencil.dim, *dims = mat->stencil.dims + 1, tmp;
1763: PetscInt *starts = mat->stencil.starts, *dxm = (PetscInt *)idxm, *dxn = (PetscInt *)idxn, sdim = dim - (1 - (PetscInt)mat->stencil.noc);
1765: PetscFunctionBegin;
1766: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
1769: PetscAssertPointer(idxm, 3);
1770: PetscAssertPointer(idxn, 5);
1772: if ((m + n) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
1773: jdxm = buf;
1774: jdxn = buf + m;
1775: } else {
1776: PetscCall(PetscMalloc2(m, &bufm, n, &bufn));
1777: jdxm = bufm;
1778: jdxn = bufn;
1779: }
1780: for (i = 0; i < m; i++) {
1781: for (j = 0; j < 3 - sdim; j++) dxm++;
1782: tmp = *dxm++ - starts[0];
1783: for (j = 0; j < dim - 1; j++) {
1784: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
1785: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
1786: }
1787: if (mat->stencil.noc) dxm++;
1788: jdxm[i] = tmp;
1789: }
1790: for (i = 0; i < n; i++) {
1791: for (j = 0; j < 3 - sdim; j++) dxn++;
1792: tmp = *dxn++ - starts[0];
1793: for (j = 0; j < dim - 1; j++) {
1794: if ((*dxn++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
1795: else tmp = tmp * dims[j] + *(dxn - 1) - starts[j + 1];
1796: }
1797: if (mat->stencil.noc) dxn++;
1798: jdxn[i] = tmp;
1799: }
1800: PetscCall(MatSetValuesLocal(mat, m, jdxm, n, jdxn, v, addv));
1801: PetscCall(PetscFree2(bufm, bufn));
1802: PetscFunctionReturn(PETSC_SUCCESS);
1803: }
1805: /*@
1806: MatSetValuesBlockedStencil - Inserts or adds a block of values into a matrix.
1807: Using structured grid indexing
1809: Not Collective
1811: Input Parameters:
1812: + mat - the matrix
1813: . m - number of rows being entered
1814: . idxm - grid coordinates for matrix rows being entered
1815: . n - number of columns being entered
1816: . idxn - grid coordinates for matrix columns being entered
1817: . v - a logically two-dimensional array of values
1818: - addv - either `ADD_VALUES` to add to existing entries or `INSERT_VALUES` to replace existing entries with new values
1820: Level: beginner
1822: Notes:
1823: By default the values, `v`, are row-oriented and unsorted.
1824: See `MatSetOption()` for other options.
1826: Calls to `MatSetValuesBlockedStencil()` with the `INSERT_VALUES` and `ADD_VALUES`
1827: options cannot be mixed without intervening calls to the assembly
1828: routines.
1830: The grid coordinates are across the entire grid, not just the local portion
1832: `MatSetValuesBlockedStencil()` uses 0-based row and column numbers in Fortran
1833: as well as in C.
1835: For setting/accessing vector values via array coordinates you can use the `DMDAVecGetArray()` routine
1837: In order to use this routine you must either obtain the matrix with `DMCreateMatrix()`
1838: or call `MatSetBlockSize()`, `MatSetLocalToGlobalMapping()` and `MatSetStencil()` first.
1840: The columns and rows in the stencil passed in MUST be contained within the
1841: ghost region of the given process as set with DMDACreateXXX() or `MatSetStencil()`. For example,
1842: if you create a `DMDA` with an overlap of one grid level and on a particular process its first
1843: local nonghost x logical coordinate is 6 (so its first ghost x logical coordinate is 5) the
1844: first i index you can use in your column and row indices in `MatSetStencil()` is 5.
1846: Negative indices may be passed in idxm and idxn, these rows and columns are
1847: simply ignored. This allows easily inserting element stiffness matrices
1848: with homogeneous Dirichlet boundary conditions that you don't want represented
1849: in the matrix.
1851: Inspired by the structured grid interface to the HYPRE package
1852: (https://computation.llnl.gov/projects/hypre-scalable-linear-solvers-multigrid-methods)
1854: Fortran Note:
1855: `idxm` and `idxn` should be declared as
1856: $ MatStencil idxm(4,m),idxn(4,n)
1857: and the values inserted using
1858: .vb
1859: idxm(MatStencil_i,1) = i
1860: idxm(MatStencil_j,1) = j
1861: idxm(MatStencil_k,1) = k
1862: etc
1863: .ve
1865: .seealso: [](ch_matrices), `Mat`, `DMDA`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`
1866: `MatSetValues()`, `MatSetValuesStencil()`, `MatSetStencil()`, `DMCreateMatrix()`, `DMDAVecGetArray()`, `MatStencil`,
1867: `MatSetBlockSize()`, `MatSetLocalToGlobalMapping()`
1868: @*/
1869: PetscErrorCode MatSetValuesBlockedStencil(Mat mat, PetscInt m, const MatStencil idxm[], PetscInt n, const MatStencil idxn[], const PetscScalar v[], InsertMode addv)
1870: {
1871: PetscInt buf[8192], *bufm = NULL, *bufn = NULL, *jdxm, *jdxn;
1872: PetscInt j, i, dim = mat->stencil.dim, *dims = mat->stencil.dims + 1, tmp;
1873: PetscInt *starts = mat->stencil.starts, *dxm = (PetscInt *)idxm, *dxn = (PetscInt *)idxn, sdim = dim - (1 - (PetscInt)mat->stencil.noc);
1875: PetscFunctionBegin;
1876: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
1879: PetscAssertPointer(idxm, 3);
1880: PetscAssertPointer(idxn, 5);
1881: PetscAssertPointer(v, 6);
1883: if ((m + n) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
1884: jdxm = buf;
1885: jdxn = buf + m;
1886: } else {
1887: PetscCall(PetscMalloc2(m, &bufm, n, &bufn));
1888: jdxm = bufm;
1889: jdxn = bufn;
1890: }
1891: for (i = 0; i < m; i++) {
1892: for (j = 0; j < 3 - sdim; j++) dxm++;
1893: tmp = *dxm++ - starts[0];
1894: for (j = 0; j < sdim - 1; j++) {
1895: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
1896: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
1897: }
1898: dxm++;
1899: jdxm[i] = tmp;
1900: }
1901: for (i = 0; i < n; i++) {
1902: for (j = 0; j < 3 - sdim; j++) dxn++;
1903: tmp = *dxn++ - starts[0];
1904: for (j = 0; j < sdim - 1; j++) {
1905: if ((*dxn++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
1906: else tmp = tmp * dims[j] + *(dxn - 1) - starts[j + 1];
1907: }
1908: dxn++;
1909: jdxn[i] = tmp;
1910: }
1911: PetscCall(MatSetValuesBlockedLocal(mat, m, jdxm, n, jdxn, v, addv));
1912: PetscCall(PetscFree2(bufm, bufn));
1913: PetscFunctionReturn(PETSC_SUCCESS);
1914: }
1916: /*@
1917: MatSetStencil - Sets the grid information for setting values into a matrix via
1918: `MatSetValuesStencil()`
1920: Not Collective
1922: Input Parameters:
1923: + mat - the matrix
1924: . dim - dimension of the grid 1, 2, or 3
1925: . dims - number of grid points in x, y, and z direction, including ghost points on your processor
1926: . starts - starting point of ghost nodes on your processor in x, y, and z direction
1927: - dof - number of degrees of freedom per node
1929: Level: beginner
1931: Notes:
1932: Inspired by the structured grid interface to the HYPRE package
1933: (www.llnl.gov/CASC/hyper)
1935: For matrices generated with `DMCreateMatrix()` this routine is automatically called and so not needed by the
1936: user.
1938: .seealso: [](ch_matrices), `Mat`, `MatStencil`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`
1939: `MatSetValues()`, `MatSetValuesBlockedStencil()`, `MatSetValuesStencil()`
1940: @*/
1941: PetscErrorCode MatSetStencil(Mat mat, PetscInt dim, const PetscInt dims[], const PetscInt starts[], PetscInt dof)
1942: {
1943: PetscFunctionBegin;
1945: PetscAssertPointer(dims, 3);
1946: PetscAssertPointer(starts, 4);
1948: mat->stencil.dim = dim + (dof > 1);
1949: for (PetscInt i = 0; i < dim; i++) {
1950: mat->stencil.dims[i] = dims[dim - i - 1]; /* copy the values in backwards */
1951: mat->stencil.starts[i] = starts[dim - i - 1];
1952: }
1953: mat->stencil.dims[dim] = dof;
1954: mat->stencil.starts[dim] = 0;
1955: mat->stencil.noc = (PetscBool)(dof == 1);
1956: PetscFunctionReturn(PETSC_SUCCESS);
1957: }
1959: /*@
1960: MatSetValuesBlocked - Inserts or adds a block of values into a matrix.
1962: Not Collective
1964: Input Parameters:
1965: + mat - the matrix
1966: . v - a logically two-dimensional array of values
1967: . m - the number of block rows
1968: . idxm - the global block indices
1969: . n - the number of block columns
1970: . idxn - the global block indices
1971: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` replaces existing entries with new values
1973: Level: intermediate
1975: Notes:
1976: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call
1977: MatXXXXSetPreallocation() or `MatSetUp()` before using this routine.
1979: The `m` and `n` count the NUMBER of blocks in the row direction and column direction,
1980: NOT the total number of rows/columns; for example, if the block size is 2 and
1981: you are passing in values for rows 2,3,4,5 then `m` would be 2 (not 4).
1982: The values in `idxm` would be 1 2; that is the first index for each block divided by
1983: the block size.
1985: You must call `MatSetBlockSize()` when constructing this matrix (before
1986: preallocating it).
1988: By default the values, `v`, are row-oriented, so the layout of
1989: `v` is the same as for `MatSetValues()`. See `MatSetOption()` for other options.
1991: Calls to `MatSetValuesBlocked()` with the `INSERT_VALUES` and `ADD_VALUES`
1992: options cannot be mixed without intervening calls to the assembly
1993: routines.
1995: `MatSetValuesBlocked()` uses 0-based row and column numbers in Fortran
1996: as well as in C.
1998: Negative indices may be passed in `idxm` and `idxn`, these rows and columns are
1999: simply ignored. This allows easily inserting element stiffness matrices
2000: with homogeneous Dirichlet boundary conditions that you don't want represented
2001: in the matrix.
2003: Each time an entry is set within a sparse matrix via `MatSetValues()`,
2004: internal searching must be done to determine where to place the
2005: data in the matrix storage space. By instead inserting blocks of
2006: entries via `MatSetValuesBlocked()`, the overhead of matrix assembly is
2007: reduced.
2009: Example:
2010: .vb
2011: Suppose m=n=2 and block size(bs) = 2 The array is
2013: 1 2 | 3 4
2014: 5 6 | 7 8
2015: - - - | - - -
2016: 9 10 | 11 12
2017: 13 14 | 15 16
2019: v[] should be passed in like
2020: v[] = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
2022: If you are not using row-oriented storage of v (that is you called MatSetOption(mat,MAT_ROW_ORIENTED,PETSC_FALSE)) then
2023: v[] = [1,5,9,13,2,6,10,14,3,7,11,15,4,8,12,16]
2024: .ve
2026: Fortran Notes:
2027: If any of `idmx`, `idxn`, and `v` are scalars pass them using, for example,
2028: .vb
2029: MatSetValuesBlocked(mat, one, [idxm], one, [idxn], [v], INSERT_VALUES)
2030: .ve
2032: If `v` is a two-dimensional array use `reshape()` to pass it as a one dimensional array
2034: .seealso: [](ch_matrices), `Mat`, `MatSetBlockSize()`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetValuesBlockedLocal()`
2035: @*/
2036: PetscErrorCode MatSetValuesBlocked(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], const PetscScalar v[], InsertMode addv)
2037: {
2038: PetscFunctionBeginHot;
2041: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2042: PetscAssertPointer(idxm, 3);
2043: PetscAssertPointer(idxn, 5);
2044: MatCheckPreallocated(mat, 1);
2045: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
2046: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
2047: if (PetscDefined(USE_DEBUG)) {
2048: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2049: PetscCheck(mat->ops->setvaluesblocked || mat->ops->setvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2050: }
2051: if (PetscDefined(USE_DEBUG)) {
2052: PetscInt rbs, cbs, M, N, i;
2053: PetscCall(MatGetBlockSizes(mat, &rbs, &cbs));
2054: PetscCall(MatGetSize(mat, &M, &N));
2055: 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);
2056: for (i = 0; i < n; i++)
2057: 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);
2058: }
2059: if (mat->assembled) {
2060: mat->was_assembled = PETSC_TRUE;
2061: mat->assembled = PETSC_FALSE;
2062: }
2063: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
2064: if (mat->ops->setvaluesblocked) {
2065: PetscUseTypeMethod(mat, setvaluesblocked, m, idxm, n, idxn, v, addv);
2066: } else {
2067: PetscInt buf[8192], *bufr = NULL, *bufc = NULL, *iidxm, *iidxn;
2068: PetscInt i, j, bs, cbs;
2070: PetscCall(MatGetBlockSizes(mat, &bs, &cbs));
2071: if ((m * bs + n * cbs) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2072: iidxm = buf;
2073: iidxn = buf + m * bs;
2074: } else {
2075: PetscCall(PetscMalloc2(m * bs, &bufr, n * cbs, &bufc));
2076: iidxm = bufr;
2077: iidxn = bufc;
2078: }
2079: for (i = 0; i < m; i++) {
2080: for (j = 0; j < bs; j++) iidxm[i * bs + j] = bs * idxm[i] + j;
2081: }
2082: if (m != n || bs != cbs || idxm != idxn) {
2083: for (i = 0; i < n; i++) {
2084: for (j = 0; j < cbs; j++) iidxn[i * cbs + j] = cbs * idxn[i] + j;
2085: }
2086: } else iidxn = iidxm;
2087: PetscCall(MatSetValues(mat, m * bs, iidxm, n * cbs, iidxn, v, addv));
2088: PetscCall(PetscFree2(bufr, bufc));
2089: }
2090: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
2091: PetscFunctionReturn(PETSC_SUCCESS);
2092: }
2094: /*@
2095: MatGetValues - Gets a block of local values from a matrix.
2097: Not Collective; can only return values that are owned by the give process
2099: Input Parameters:
2100: + mat - the matrix
2101: . v - a logically two-dimensional array for storing the values
2102: . m - the number of rows
2103: . idxm - the global indices of the rows
2104: . n - the number of columns
2105: - idxn - the global indices of the columns
2107: Level: advanced
2109: Notes:
2110: The user must allocate space (m*n `PetscScalar`s) for the values, `v`.
2111: The values, `v`, are then returned in a row-oriented format,
2112: analogous to that used by default in `MatSetValues()`.
2114: `MatGetValues()` uses 0-based row and column numbers in
2115: Fortran as well as in C.
2117: `MatGetValues()` requires that the matrix has been assembled
2118: with `MatAssemblyBegin()`/`MatAssemblyEnd()`. Thus, calls to
2119: `MatSetValues()` and `MatGetValues()` CANNOT be made in succession
2120: without intermediate matrix assembly.
2122: Negative row or column indices will be ignored and those locations in `v` will be
2123: left unchanged.
2125: For the standard row-based matrix formats, `idxm` can only contain rows owned by the requesting MPI process.
2126: That is, rows with global index greater than or equal to rstart and less than rend where rstart and rend are obtainable
2127: from `MatGetOwnershipRange`(mat,&rstart,&rend).
2129: .seealso: [](ch_matrices), `Mat`, `MatGetRow()`, `MatCreateSubMatrices()`, `MatSetValues()`, `MatGetOwnershipRange()`, `MatGetValuesLocal()`, `MatGetValue()`
2130: @*/
2131: PetscErrorCode MatGetValues(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], PetscScalar v[])
2132: {
2133: PetscFunctionBegin;
2136: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS);
2137: PetscAssertPointer(idxm, 3);
2138: PetscAssertPointer(idxn, 5);
2139: PetscAssertPointer(v, 6);
2140: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2141: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2142: MatCheckPreallocated(mat, 1);
2144: PetscCall(PetscLogEventBegin(MAT_GetValues, mat, 0, 0, 0));
2145: PetscUseTypeMethod(mat, getvalues, m, idxm, n, idxn, v);
2146: PetscCall(PetscLogEventEnd(MAT_GetValues, mat, 0, 0, 0));
2147: PetscFunctionReturn(PETSC_SUCCESS);
2148: }
2150: /*@
2151: MatGetValuesLocal - retrieves values from certain locations in a matrix using the local numbering of the indices
2152: defined previously by `MatSetLocalToGlobalMapping()`
2154: Not Collective
2156: Input Parameters:
2157: + mat - the matrix
2158: . nrow - number of rows
2159: . irow - the row local indices
2160: . ncol - number of columns
2161: - icol - the column local indices
2163: Output Parameter:
2164: . y - a logically two-dimensional array of values
2166: Level: advanced
2168: Notes:
2169: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call `MatSetLocalToGlobalMapping()` before using this routine.
2171: 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,
2172: are greater than or equal to rstart and less than rend where rstart and rend are obtainable from `MatGetOwnershipRange`(mat,&rstart,&rend). One can
2173: determine if the resulting global row associated with the local row r is owned by the requesting MPI process by applying the `ISLocalToGlobalMapping` set
2174: with `MatSetLocalToGlobalMapping()`.
2176: Developer Note:
2177: This is labelled with C so does not automatically generate Fortran stubs and interfaces
2178: because it requires multiple Fortran interfaces depending on which arguments are scalar or arrays.
2180: .seealso: [](ch_matrices), `Mat`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetLocalToGlobalMapping()`,
2181: `MatSetValuesLocal()`, `MatGetValues()`
2182: @*/
2183: PetscErrorCode MatGetValuesLocal(Mat mat, PetscInt nrow, const PetscInt irow[], PetscInt ncol, const PetscInt icol[], PetscScalar y[])
2184: {
2185: PetscFunctionBeginHot;
2188: MatCheckPreallocated(mat, 1);
2189: if (!nrow || !ncol) PetscFunctionReturn(PETSC_SUCCESS); /* no values to retrieve */
2190: PetscAssertPointer(irow, 3);
2191: PetscAssertPointer(icol, 5);
2192: if (PetscDefined(USE_DEBUG)) {
2193: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2194: PetscCheck(mat->ops->getvalueslocal || mat->ops->getvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2195: }
2196: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2197: PetscCall(PetscLogEventBegin(MAT_GetValues, mat, 0, 0, 0));
2198: if (mat->ops->getvalueslocal) PetscUseTypeMethod(mat, getvalueslocal, nrow, irow, ncol, icol, y);
2199: else {
2200: PetscInt buf[8192], *bufr = NULL, *bufc = NULL, *irowm, *icolm;
2201: if ((nrow + ncol) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2202: irowm = buf;
2203: icolm = buf + nrow;
2204: } else {
2205: PetscCall(PetscMalloc2(nrow, &bufr, ncol, &bufc));
2206: irowm = bufr;
2207: icolm = bufc;
2208: }
2209: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "MatGetValuesLocal() cannot proceed without local-to-global row mapping (See MatSetLocalToGlobalMapping()).");
2210: PetscCheck(mat->cmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "MatGetValuesLocal() cannot proceed without local-to-global column mapping (See MatSetLocalToGlobalMapping()).");
2211: PetscCall(ISLocalToGlobalMappingApply(mat->rmap->mapping, nrow, irow, irowm));
2212: PetscCall(ISLocalToGlobalMappingApply(mat->cmap->mapping, ncol, icol, icolm));
2213: PetscCall(MatGetValues(mat, nrow, irowm, ncol, icolm, y));
2214: PetscCall(PetscFree2(bufr, bufc));
2215: }
2216: PetscCall(PetscLogEventEnd(MAT_GetValues, mat, 0, 0, 0));
2217: PetscFunctionReturn(PETSC_SUCCESS);
2218: }
2220: /*@
2221: MatSetValuesBatch - Adds (`ADD_VALUES`) many blocks of values into a matrix at once. The blocks must all be square and
2222: the same size. Currently, this can only be called once and creates the given matrix.
2224: Not Collective
2226: Input Parameters:
2227: + mat - the matrix
2228: . nb - the number of blocks
2229: . bs - the number of rows (and columns) in each block
2230: . rows - a concatenation of the rows for each block
2231: - v - a concatenation of logically two-dimensional arrays of values
2233: Level: advanced
2235: Notes:
2236: `MatSetPreallocationCOO()` and `MatSetValuesCOO()` may be a better way to provide the values
2238: In the future, we will extend this routine to handle rectangular blocks, and to allow multiple calls for a given matrix.
2240: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
2241: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`, `MatSetValues()`, `MatSetPreallocationCOO()`, `MatSetValuesCOO()`
2242: @*/
2243: PetscErrorCode MatSetValuesBatch(Mat mat, PetscInt nb, PetscInt bs, PetscInt rows[], const PetscScalar v[])
2244: {
2245: PetscFunctionBegin;
2248: PetscAssertPointer(rows, 4);
2249: PetscAssertPointer(v, 5);
2250: PetscAssert(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2252: PetscCall(PetscLogEventBegin(MAT_SetValuesBatch, mat, 0, 0, 0));
2253: if (mat->ops->setvaluesbatch) PetscUseTypeMethod(mat, setvaluesbatch, nb, bs, rows, v);
2254: else {
2255: for (PetscInt b = 0; b < nb; ++b) PetscCall(MatSetValues(mat, bs, &rows[b * bs], bs, &rows[b * bs], &v[b * bs * bs], ADD_VALUES));
2256: }
2257: PetscCall(PetscLogEventEnd(MAT_SetValuesBatch, mat, 0, 0, 0));
2258: PetscFunctionReturn(PETSC_SUCCESS);
2259: }
2261: /*@
2262: MatSetLocalToGlobalMapping - Sets a local-to-global numbering for use by
2263: the routine `MatSetValuesLocal()` to allow users to insert matrix entries
2264: using a local (per-processor) numbering.
2266: Not Collective
2268: Input Parameters:
2269: + x - the matrix
2270: . rmapping - row mapping created with `ISLocalToGlobalMappingCreate()` or `ISLocalToGlobalMappingCreateIS()`
2271: - cmapping - column mapping
2273: Level: intermediate
2275: Note:
2276: If the matrix is obtained with `DMCreateMatrix()` then this may already have been called on the matrix
2278: .seealso: [](ch_matrices), `Mat`, `DM`, `DMCreateMatrix()`, `MatGetLocalToGlobalMapping()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetValuesLocal()`, `MatGetValuesLocal()`
2279: @*/
2280: PetscErrorCode MatSetLocalToGlobalMapping(Mat x, ISLocalToGlobalMapping rmapping, ISLocalToGlobalMapping cmapping)
2281: {
2282: PetscFunctionBegin;
2287: if (x->ops->setlocaltoglobalmapping) PetscUseTypeMethod(x, setlocaltoglobalmapping, rmapping, cmapping);
2288: else {
2289: PetscCall(PetscLayoutSetISLocalToGlobalMapping(x->rmap, rmapping));
2290: PetscCall(PetscLayoutSetISLocalToGlobalMapping(x->cmap, cmapping));
2291: }
2292: PetscFunctionReturn(PETSC_SUCCESS);
2293: }
2295: /*@
2296: MatGetLocalToGlobalMapping - Gets the local-to-global numbering set by `MatSetLocalToGlobalMapping()`
2298: Not Collective
2300: Input Parameter:
2301: . A - the matrix
2303: Output Parameters:
2304: + rmapping - row mapping
2305: - cmapping - column mapping
2307: Level: advanced
2309: .seealso: [](ch_matrices), `Mat`, `MatSetLocalToGlobalMapping()`, `MatSetValuesLocal()`
2310: @*/
2311: PetscErrorCode MatGetLocalToGlobalMapping(Mat A, ISLocalToGlobalMapping *rmapping, ISLocalToGlobalMapping *cmapping)
2312: {
2313: PetscFunctionBegin;
2316: if (rmapping) {
2317: PetscAssertPointer(rmapping, 2);
2318: *rmapping = A->rmap->mapping;
2319: }
2320: if (cmapping) {
2321: PetscAssertPointer(cmapping, 3);
2322: *cmapping = A->cmap->mapping;
2323: }
2324: PetscFunctionReturn(PETSC_SUCCESS);
2325: }
2327: /*@
2328: MatSetLayouts - Sets the `PetscLayout` objects for rows and columns of a matrix
2330: Logically Collective
2332: Input Parameters:
2333: + A - the matrix
2334: . rmap - row layout
2335: - cmap - column layout
2337: Level: advanced
2339: Note:
2340: The `PetscLayout` objects are usually created automatically for the matrix so this routine rarely needs to be called.
2342: .seealso: [](ch_matrices), `Mat`, `PetscLayout`, `MatCreateVecs()`, `MatGetLocalToGlobalMapping()`, `MatGetLayouts()`
2343: @*/
2344: PetscErrorCode MatSetLayouts(Mat A, PetscLayout rmap, PetscLayout cmap)
2345: {
2346: PetscFunctionBegin;
2348: PetscCall(PetscLayoutReference(rmap, &A->rmap));
2349: PetscCall(PetscLayoutReference(cmap, &A->cmap));
2350: PetscFunctionReturn(PETSC_SUCCESS);
2351: }
2353: /*@
2354: MatGetLayouts - Gets the `PetscLayout` objects for rows and columns
2356: Not Collective
2358: Input Parameter:
2359: . A - the matrix
2361: Output Parameters:
2362: + rmap - row layout
2363: - cmap - column layout
2365: Level: advanced
2367: .seealso: [](ch_matrices), `Mat`, [Matrix Layouts](sec_matlayout), `PetscLayout`, `MatCreateVecs()`, `MatGetLocalToGlobalMapping()`, `MatSetLayouts()`
2368: @*/
2369: PetscErrorCode MatGetLayouts(Mat A, PetscLayout *rmap, PetscLayout *cmap)
2370: {
2371: PetscFunctionBegin;
2374: if (rmap) {
2375: PetscAssertPointer(rmap, 2);
2376: *rmap = A->rmap;
2377: }
2378: if (cmap) {
2379: PetscAssertPointer(cmap, 3);
2380: *cmap = A->cmap;
2381: }
2382: PetscFunctionReturn(PETSC_SUCCESS);
2383: }
2385: /*@
2386: MatSetValuesLocal - Inserts or adds values into certain locations of a matrix,
2387: using a local numbering of the rows and columns.
2389: Not Collective
2391: Input Parameters:
2392: + mat - the matrix
2393: . nrow - number of rows
2394: . irow - the row local indices
2395: . ncol - number of columns
2396: . icol - the column local indices
2397: . y - a logically two-dimensional array of values
2398: - addv - either `INSERT_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
2400: Level: intermediate
2402: Notes:
2403: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call `MatSetLocalToGlobalMapping()` before using this routine
2405: Calls to `MatSetValuesLocal()` with the `INSERT_VALUES` and `ADD_VALUES`
2406: options cannot be mixed without intervening calls to the assembly
2407: routines.
2409: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
2410: MUST be called after all calls to `MatSetValuesLocal()` have been completed.
2412: Fortran Notes:
2413: If any of `irow`, `icol`, and `y` are scalars pass them using, for example,
2414: .vb
2415: MatSetValuesLocal(mat, one, [irow], one, [icol], [y], INSERT_VALUES)
2416: .ve
2418: If `y` is a two-dimensional array use `reshape()` to pass it as a one dimensional array
2420: Developer Note:
2421: This is labeled with C so does not automatically generate Fortran stubs and interfaces
2422: because it requires multiple Fortran interfaces depending on which arguments are scalar or arrays.
2424: .seealso: [](ch_matrices), `Mat`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetLocalToGlobalMapping()`,
2425: `MatGetValuesLocal()`
2426: @*/
2427: PetscErrorCode MatSetValuesLocal(Mat mat, PetscInt nrow, const PetscInt irow[], PetscInt ncol, const PetscInt icol[], const PetscScalar y[], InsertMode addv)
2428: {
2429: PetscFunctionBeginHot;
2432: MatCheckPreallocated(mat, 1);
2433: if (!nrow || !ncol) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2434: PetscAssertPointer(irow, 3);
2435: PetscAssertPointer(icol, 5);
2436: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
2437: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
2438: if (PetscDefined(USE_DEBUG)) {
2439: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2440: PetscCheck(mat->ops->setvalueslocal || mat->ops->setvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2441: }
2443: if (mat->assembled) {
2444: mat->was_assembled = PETSC_TRUE;
2445: mat->assembled = PETSC_FALSE;
2446: }
2447: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
2448: if (mat->ops->setvalueslocal) PetscUseTypeMethod(mat, setvalueslocal, nrow, irow, ncol, icol, y, addv);
2449: else {
2450: PetscInt buf[8192], *bufr = NULL, *bufc = NULL;
2451: const PetscInt *irowm, *icolm;
2453: if ((!mat->rmap->mapping && !mat->cmap->mapping) || (nrow + ncol) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2454: bufr = buf;
2455: bufc = buf + nrow;
2456: irowm = bufr;
2457: icolm = bufc;
2458: } else {
2459: PetscCall(PetscMalloc2(nrow, &bufr, ncol, &bufc));
2460: irowm = bufr;
2461: icolm = bufc;
2462: }
2463: if (mat->rmap->mapping) PetscCall(ISLocalToGlobalMappingApply(mat->rmap->mapping, nrow, irow, bufr));
2464: else irowm = irow;
2465: if (mat->cmap->mapping) {
2466: if (mat->cmap->mapping != mat->rmap->mapping || ncol != nrow || icol != irow) {
2467: PetscCall(ISLocalToGlobalMappingApply(mat->cmap->mapping, ncol, icol, bufc));
2468: } else icolm = irowm;
2469: } else icolm = icol;
2470: PetscCall(MatSetValues(mat, nrow, irowm, ncol, icolm, y, addv));
2471: if (bufr != buf) PetscCall(PetscFree2(bufr, bufc));
2472: }
2473: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
2474: PetscFunctionReturn(PETSC_SUCCESS);
2475: }
2477: /*@
2478: MatSetValuesBlockedLocal - Inserts or adds values into certain locations of a matrix,
2479: using a local ordering of the nodes a block at a time.
2481: Not Collective
2483: Input Parameters:
2484: + mat - the matrix
2485: . nrow - number of rows
2486: . irow - the row local indices
2487: . ncol - number of columns
2488: . icol - the column local indices
2489: . y - a logically two-dimensional array of values
2490: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
2492: Level: intermediate
2494: Notes:
2495: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call `MatSetBlockSize()` and `MatSetLocalToGlobalMapping()`
2496: before using this routineBefore calling `MatSetValuesLocal()`, the user must first set the
2498: Calls to `MatSetValuesBlockedLocal()` with the `INSERT_VALUES` and `ADD_VALUES`
2499: options cannot be mixed without intervening calls to the assembly
2500: routines.
2502: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
2503: MUST be called after all calls to `MatSetValuesBlockedLocal()` have been completed.
2505: Fortran Notes:
2506: If any of `irow`, `icol`, and `y` are scalars pass them using, for example,
2507: .vb
2508: MatSetValuesBlockedLocal(mat, one, [irow], one, [icol], [y], INSERT_VALUES)
2509: .ve
2511: If `y` is a two-dimensional array use `reshape()` to pass it as a one dimensional array
2513: Developer Note:
2514: This is labeled with C so does not automatically generate Fortran stubs and interfaces
2515: because it requires multiple Fortran interfaces depending on which arguments are scalar or arrays.
2517: .seealso: [](ch_matrices), `Mat`, `MatSetBlockSize()`, `MatSetLocalToGlobalMapping()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`,
2518: `MatSetValuesLocal()`, `MatSetValuesBlocked()`
2519: @*/
2520: PetscErrorCode MatSetValuesBlockedLocal(Mat mat, PetscInt nrow, const PetscInt irow[], PetscInt ncol, const PetscInt icol[], const PetscScalar y[], InsertMode addv)
2521: {
2522: PetscFunctionBeginHot;
2525: MatCheckPreallocated(mat, 1);
2526: if (!nrow || !ncol) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2527: PetscAssertPointer(irow, 3);
2528: PetscAssertPointer(icol, 5);
2529: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
2530: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
2531: if (PetscDefined(USE_DEBUG)) {
2532: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2533: 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);
2534: }
2536: if (mat->assembled) {
2537: mat->was_assembled = PETSC_TRUE;
2538: mat->assembled = PETSC_FALSE;
2539: }
2540: if (PetscUnlikelyDebug(mat->rmap->mapping)) { /* Condition on the mapping existing, because MatSetValuesBlockedLocal_IS does not require it to be set. */
2541: PetscInt irbs, rbs;
2542: PetscCall(MatGetBlockSizes(mat, &rbs, NULL));
2543: PetscCall(ISLocalToGlobalMappingGetBlockSize(mat->rmap->mapping, &irbs));
2544: PetscCheck(rbs == irbs, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Different row block sizes! mat %" PetscInt_FMT ", row l2g map %" PetscInt_FMT, rbs, irbs);
2545: }
2546: if (PetscUnlikelyDebug(mat->cmap->mapping)) {
2547: PetscInt icbs, cbs;
2548: PetscCall(MatGetBlockSizes(mat, NULL, &cbs));
2549: PetscCall(ISLocalToGlobalMappingGetBlockSize(mat->cmap->mapping, &icbs));
2550: PetscCheck(cbs == icbs, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Different col block sizes! mat %" PetscInt_FMT ", col l2g map %" PetscInt_FMT, cbs, icbs);
2551: }
2552: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
2553: if (mat->ops->setvaluesblockedlocal) PetscUseTypeMethod(mat, setvaluesblockedlocal, nrow, irow, ncol, icol, y, addv);
2554: else {
2555: PetscInt buf[8192], *bufr = NULL, *bufc = NULL;
2556: const PetscInt *irowm, *icolm;
2558: if ((!mat->rmap->mapping && !mat->cmap->mapping) || (nrow + ncol) <= ((PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf))) {
2559: bufr = buf;
2560: bufc = buf + nrow;
2561: irowm = bufr;
2562: icolm = bufc;
2563: } else {
2564: PetscCall(PetscMalloc2(nrow, &bufr, ncol, &bufc));
2565: irowm = bufr;
2566: icolm = bufc;
2567: }
2568: if (mat->rmap->mapping) PetscCall(ISLocalToGlobalMappingApplyBlock(mat->rmap->mapping, nrow, irow, bufr));
2569: else irowm = irow;
2570: if (mat->cmap->mapping) {
2571: if (mat->cmap->mapping != mat->rmap->mapping || ncol != nrow || icol != irow) {
2572: PetscCall(ISLocalToGlobalMappingApplyBlock(mat->cmap->mapping, ncol, icol, bufc));
2573: } else icolm = irowm;
2574: } else icolm = icol;
2575: PetscCall(MatSetValuesBlocked(mat, nrow, irowm, ncol, icolm, y, addv));
2576: if (bufr != buf) PetscCall(PetscFree2(bufr, bufc));
2577: }
2578: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
2579: PetscFunctionReturn(PETSC_SUCCESS);
2580: }
2582: /*@
2583: MatMultDiagonalBlock - Computes the matrix-vector product, $y = Dx$. Where `D` is defined by the inode or block structure of the diagonal
2585: Collective
2587: Input Parameters:
2588: + mat - the matrix
2589: - x - the vector to be multiplied
2591: Output Parameter:
2592: . y - the result
2594: Level: developer
2596: Note:
2597: The vectors `x` and `y` cannot be the same. I.e., one cannot
2598: call `MatMultDiagonalBlock`(A,y,y).
2600: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
2601: @*/
2602: PetscErrorCode MatMultDiagonalBlock(Mat mat, Vec x, Vec y)
2603: {
2604: PetscFunctionBegin;
2610: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2611: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2612: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2613: MatCheckPreallocated(mat, 1);
2615: PetscUseTypeMethod(mat, multdiagonalblock, x, y);
2616: PetscCall(PetscObjectStateIncrease((PetscObject)y));
2617: PetscFunctionReturn(PETSC_SUCCESS);
2618: }
2620: /*@
2621: MatMult - Computes the matrix-vector product, $y = Ax$.
2623: Neighbor-wise Collective
2625: Input Parameters:
2626: + mat - the matrix
2627: - x - the vector to be multiplied
2629: Output Parameter:
2630: . y - the result
2632: Level: beginner
2634: Note:
2635: The vectors `x` and `y` cannot be the same. I.e., one cannot
2636: call `MatMult`(A,y,y).
2638: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
2639: @*/
2640: PetscErrorCode MatMult(Mat mat, Vec x, Vec y)
2641: {
2642: PetscFunctionBegin;
2646: VecCheckAssembled(x);
2648: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2649: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2650: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2651: 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);
2652: 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);
2653: 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);
2654: 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);
2655: PetscCall(VecSetErrorIfLocked(y, 3));
2656: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
2657: MatCheckPreallocated(mat, 1);
2659: PetscCall(VecLockReadPush(x));
2660: PetscCall(PetscLogEventBegin(MAT_Mult, mat, x, y, 0));
2661: PetscUseTypeMethod(mat, mult, x, y);
2662: PetscCall(PetscLogEventEnd(MAT_Mult, mat, x, y, 0));
2663: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
2664: PetscCall(VecLockReadPop(x));
2665: PetscFunctionReturn(PETSC_SUCCESS);
2666: }
2668: /*@
2669: MatMultTranspose - Computes matrix transpose times a vector $y = A^T * x$.
2671: Neighbor-wise Collective
2673: Input Parameters:
2674: + mat - the matrix
2675: - x - the vector to be multiplied
2677: Output Parameter:
2678: . y - the result
2680: Level: beginner
2682: Notes:
2683: The vectors `x` and `y` cannot be the same. I.e., one cannot
2684: call `MatMultTranspose`(A,y,y).
2686: For complex numbers this does NOT compute the Hermitian (complex conjugate) transpose multiple,
2687: use `MatMultHermitianTranspose()`
2689: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatMultHermitianTranspose()`, `MatTranspose()`
2690: @*/
2691: PetscErrorCode MatMultTranspose(Mat mat, Vec x, Vec y)
2692: {
2693: PetscErrorCode (*op)(Mat, Vec, Vec) = NULL;
2695: PetscFunctionBegin;
2699: VecCheckAssembled(x);
2702: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2703: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2704: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2705: 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);
2706: 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);
2707: 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);
2708: 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);
2709: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
2710: MatCheckPreallocated(mat, 1);
2712: if (!mat->ops->multtranspose) {
2713: if (mat->symmetric == PETSC_BOOL3_TRUE && mat->ops->mult) op = mat->ops->mult;
2714: 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);
2715: } else op = mat->ops->multtranspose;
2716: PetscCall(PetscLogEventBegin(MAT_MultTranspose, mat, x, y, 0));
2717: PetscCall(VecLockReadPush(x));
2718: PetscCall((*op)(mat, x, y));
2719: PetscCall(VecLockReadPop(x));
2720: PetscCall(PetscLogEventEnd(MAT_MultTranspose, mat, x, y, 0));
2721: PetscCall(PetscObjectStateIncrease((PetscObject)y));
2722: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
2723: PetscFunctionReturn(PETSC_SUCCESS);
2724: }
2726: /*@
2727: MatMultHermitianTranspose - Computes matrix Hermitian-transpose times a vector $y = A^H * x$.
2729: Neighbor-wise Collective
2731: Input Parameters:
2732: + mat - the matrix
2733: - x - the vector to be multiplied
2735: Output Parameter:
2736: . y - the result
2738: Level: beginner
2740: Notes:
2741: The vectors `x` and `y` cannot be the same. I.e., one cannot
2742: call `MatMultHermitianTranspose`(A,y,y).
2744: Also called the conjugate transpose, complex conjugate transpose, or adjoint.
2746: For real numbers `MatMultTranspose()` and `MatMultHermitianTranspose()` are identical.
2748: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultAdd()`, `MatMultHermitianTransposeAdd()`, `MatMultTranspose()`
2749: @*/
2750: PetscErrorCode MatMultHermitianTranspose(Mat mat, Vec x, Vec y)
2751: {
2752: PetscFunctionBegin;
2758: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2759: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2760: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2761: 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);
2762: 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);
2763: 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);
2764: 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);
2765: MatCheckPreallocated(mat, 1);
2767: PetscCall(PetscLogEventBegin(MAT_MultHermitianTranspose, mat, x, y, 0));
2768: #if defined(PETSC_USE_COMPLEX)
2769: if (mat->ops->multhermitiantranspose || (mat->hermitian == PETSC_BOOL3_TRUE && mat->ops->mult)) {
2770: PetscCall(VecLockReadPush(x));
2771: if (mat->ops->multhermitiantranspose) PetscUseTypeMethod(mat, multhermitiantranspose, x, y);
2772: else PetscUseTypeMethod(mat, mult, x, y);
2773: PetscCall(VecLockReadPop(x));
2774: } else {
2775: Vec w;
2776: PetscCall(VecDuplicate(x, &w));
2777: PetscCall(VecCopy(x, w));
2778: PetscCall(VecConjugate(w));
2779: PetscCall(MatMultTranspose(mat, w, y));
2780: PetscCall(VecDestroy(&w));
2781: PetscCall(VecConjugate(y));
2782: }
2783: PetscCall(PetscObjectStateIncrease((PetscObject)y));
2784: #else
2785: PetscCall(MatMultTranspose(mat, x, y));
2786: #endif
2787: PetscCall(PetscLogEventEnd(MAT_MultHermitianTranspose, mat, x, y, 0));
2788: PetscFunctionReturn(PETSC_SUCCESS);
2789: }
2791: /*@
2792: MatMultAdd - Computes $v3 = v2 + A * v1$.
2794: Neighbor-wise Collective
2796: Input Parameters:
2797: + mat - the matrix
2798: . v1 - the vector to be multiplied by `mat`
2799: - v2 - the vector to be added to the result
2801: Output Parameter:
2802: . v3 - the result
2804: Level: beginner
2806: Note:
2807: The vectors `v1` and `v3` cannot be the same. I.e., one cannot
2808: call `MatMultAdd`(A,v1,v2,v1).
2810: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMult()`, `MatMultTransposeAdd()`
2811: @*/
2812: PetscErrorCode MatMultAdd(Mat mat, Vec v1, Vec v2, Vec v3)
2813: {
2814: PetscFunctionBegin;
2821: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2822: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2823: 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);
2824: /* 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);
2825: 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); */
2826: 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);
2827: 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);
2828: PetscCheck(v1 != v3, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "v1 and v3 must be different vectors");
2829: MatCheckPreallocated(mat, 1);
2831: PetscCall(PetscLogEventBegin(MAT_MultAdd, mat, v1, v2, v3));
2832: PetscCall(VecLockReadPush(v1));
2833: PetscUseTypeMethod(mat, multadd, v1, v2, v3);
2834: PetscCall(VecLockReadPop(v1));
2835: PetscCall(PetscLogEventEnd(MAT_MultAdd, mat, v1, v2, v3));
2836: PetscCall(PetscObjectStateIncrease((PetscObject)v3));
2837: PetscFunctionReturn(PETSC_SUCCESS);
2838: }
2840: /*@
2841: MatMultTransposeAdd - Computes $v3 = v2 + A^T * v1$.
2843: Neighbor-wise Collective
2845: Input Parameters:
2846: + mat - the matrix
2847: . v1 - the vector to be multiplied by the transpose of the matrix
2848: - v2 - the vector to be added to the result
2850: Output Parameter:
2851: . v3 - the result
2853: Level: beginner
2855: Note:
2856: The vectors `v1` and `v3` cannot be the same. I.e., one cannot
2857: call `MatMultTransposeAdd`(A,v1,v2,v1).
2859: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMult()`
2860: @*/
2861: PetscErrorCode MatMultTransposeAdd(Mat mat, Vec v1, Vec v2, Vec v3)
2862: {
2863: PetscErrorCode (*op)(Mat, Vec, Vec, Vec) = (!mat->ops->multtransposeadd && mat->symmetric) ? mat->ops->multadd : mat->ops->multtransposeadd;
2865: PetscFunctionBegin;
2872: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2873: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2874: 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);
2875: 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);
2876: 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);
2877: PetscCheck(v1 != v3, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "v1 and v3 must be different vectors");
2878: PetscCheck(op, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2879: MatCheckPreallocated(mat, 1);
2881: PetscCall(PetscLogEventBegin(MAT_MultTransposeAdd, mat, v1, v2, v3));
2882: PetscCall(VecLockReadPush(v1));
2883: PetscCall((*op)(mat, v1, v2, v3));
2884: PetscCall(VecLockReadPop(v1));
2885: PetscCall(PetscLogEventEnd(MAT_MultTransposeAdd, mat, v1, v2, v3));
2886: PetscCall(PetscObjectStateIncrease((PetscObject)v3));
2887: PetscFunctionReturn(PETSC_SUCCESS);
2888: }
2890: /*@
2891: MatMultHermitianTransposeAdd - Computes $v3 = v2 + A^H * v1$.
2893: Neighbor-wise Collective
2895: Input Parameters:
2896: + mat - the matrix
2897: . v1 - the vector to be multiplied by the Hermitian transpose
2898: - v2 - the vector to be added to the result
2900: Output Parameter:
2901: . v3 - the result
2903: Level: beginner
2905: Note:
2906: The vectors `v1` and `v3` cannot be the same. I.e., one cannot
2907: call `MatMultHermitianTransposeAdd`(A,v1,v2,v1).
2909: .seealso: [](ch_matrices), `Mat`, `MatMultHermitianTranspose()`, `MatMultTranspose()`, `MatMultAdd()`, `MatMult()`
2910: @*/
2911: PetscErrorCode MatMultHermitianTransposeAdd(Mat mat, Vec v1, Vec v2, Vec v3)
2912: {
2913: PetscFunctionBegin;
2920: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2921: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2922: PetscCheck(v1 != v3, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "v1 and v3 must be different vectors");
2923: 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);
2924: 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);
2925: 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);
2926: MatCheckPreallocated(mat, 1);
2928: PetscCall(PetscLogEventBegin(MAT_MultHermitianTransposeAdd, mat, v1, v2, v3));
2929: PetscCall(VecLockReadPush(v1));
2930: if (mat->ops->multhermitiantransposeadd) PetscUseTypeMethod(mat, multhermitiantransposeadd, v1, v2, v3);
2931: else {
2932: Vec w, z;
2933: PetscCall(VecDuplicate(v1, &w));
2934: PetscCall(VecCopy(v1, w));
2935: PetscCall(VecConjugate(w));
2936: PetscCall(VecDuplicate(v3, &z));
2937: PetscCall(MatMultTranspose(mat, w, z));
2938: PetscCall(VecDestroy(&w));
2939: PetscCall(VecConjugate(z));
2940: if (v2 != v3) {
2941: PetscCall(VecWAXPY(v3, 1.0, v2, z));
2942: } else {
2943: PetscCall(VecAXPY(v3, 1.0, z));
2944: }
2945: PetscCall(VecDestroy(&z));
2946: }
2947: PetscCall(VecLockReadPop(v1));
2948: PetscCall(PetscLogEventEnd(MAT_MultHermitianTransposeAdd, mat, v1, v2, v3));
2949: PetscCall(PetscObjectStateIncrease((PetscObject)v3));
2950: PetscFunctionReturn(PETSC_SUCCESS);
2951: }
2953: /*@
2954: MatGetFactorType - gets the type of factorization a matrix is
2956: Not Collective
2958: Input Parameter:
2959: . mat - the matrix
2961: Output Parameter:
2962: . 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`
2964: Level: intermediate
2966: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorType`, `MatGetFactor()`, `MatSetFactorType()`, `MAT_FACTOR_NONE`, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ILU`,
2967: `MAT_FACTOR_ICC`,`MAT_FACTOR_ILUDT`, `MAT_FACTOR_QR`
2968: @*/
2969: PetscErrorCode MatGetFactorType(Mat mat, MatFactorType *t)
2970: {
2971: PetscFunctionBegin;
2974: PetscAssertPointer(t, 2);
2975: *t = mat->factortype;
2976: PetscFunctionReturn(PETSC_SUCCESS);
2977: }
2979: /*@
2980: MatSetFactorType - sets the type of factorization a matrix is
2982: Logically Collective
2984: Input Parameters:
2985: + mat - the matrix
2986: - 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`
2988: Level: intermediate
2990: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorType`, `MatGetFactor()`, `MatGetFactorType()`, `MAT_FACTOR_NONE`, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ILU`,
2991: `MAT_FACTOR_ICC`,`MAT_FACTOR_ILUDT`, `MAT_FACTOR_QR`
2992: @*/
2993: PetscErrorCode MatSetFactorType(Mat mat, MatFactorType t)
2994: {
2995: PetscFunctionBegin;
2998: mat->factortype = t;
2999: PetscFunctionReturn(PETSC_SUCCESS);
3000: }
3002: /*@
3003: MatGetInfo - Returns information about matrix storage (number of
3004: nonzeros, memory, etc.).
3006: Collective if `MAT_GLOBAL_MAX` or `MAT_GLOBAL_SUM` is used as the flag
3008: Input Parameters:
3009: + mat - the matrix
3010: - flag - flag indicating the type of parameters to be returned (`MAT_LOCAL` - local matrix, `MAT_GLOBAL_MAX` - maximum over all processors, `MAT_GLOBAL_SUM` - sum over all processors)
3012: Output Parameter:
3013: . info - matrix information context
3015: Options Database Key:
3016: . -mat_view ::ascii_info - print matrix info to `PETSC_STDOUT`
3018: Level: intermediate
3020: Notes:
3021: The `MatInfo` context contains a variety of matrix data, including
3022: number of nonzeros allocated and used, number of mallocs during
3023: matrix assembly, etc. Additional information for factored matrices
3024: is provided (such as the fill ratio, number of mallocs during
3025: factorization, etc.).
3027: Example:
3028: See the file ${PETSC_DIR}/include/petscmat.h for a complete list of
3029: data within the `MatInfo` context. For example,
3030: .vb
3031: MatInfo info;
3032: Mat A;
3033: double mal, nz_a, nz_u;
3035: MatGetInfo(A, MAT_LOCAL, &info);
3036: mal = info.mallocs;
3037: nz_a = info.nz_allocated;
3038: .ve
3040: Fortran Note:
3041: Declare info as a `MatInfo` array of dimension `MAT_INFO_SIZE`, and then extract the parameters
3042: of interest. See the file ${PETSC_DIR}/include/petsc/finclude/petscmat.h
3043: a complete list of parameter names.
3044: .vb
3045: MatInfo info(MAT_INFO_SIZE)
3046: double precision mal, nz_a
3047: Mat A
3048: integer ierr
3050: call MatGetInfo(A, MAT_LOCAL, info, ierr)
3051: mal = info(MAT_INFO_MALLOCS)
3052: nz_a = info(MAT_INFO_NZ_ALLOCATED)
3053: .ve
3055: .seealso: [](ch_matrices), `Mat`, `MatInfo`, `MatStashGetInfo()`
3056: @*/
3057: PetscErrorCode MatGetInfo(Mat mat, MatInfoType flag, MatInfo *info)
3058: {
3059: PetscFunctionBegin;
3062: PetscAssertPointer(info, 3);
3063: MatCheckPreallocated(mat, 1);
3064: PetscUseTypeMethod(mat, getinfo, flag, info);
3065: PetscFunctionReturn(PETSC_SUCCESS);
3066: }
3068: /*
3069: This is used by external packages where it is not easy to get the info from the actual
3070: matrix factorization.
3071: */
3072: PetscErrorCode MatGetInfo_External(Mat A, MatInfoType flag, MatInfo *info)
3073: {
3074: PetscFunctionBegin;
3075: PetscCall(PetscMemzero(info, sizeof(MatInfo)));
3076: PetscFunctionReturn(PETSC_SUCCESS);
3077: }
3079: /*@
3080: MatLUFactor - Performs in-place LU factorization of matrix.
3082: Collective
3084: Input Parameters:
3085: + mat - the matrix
3086: . row - row permutation
3087: . col - column permutation
3088: - info - options for factorization, includes
3089: .vb
3090: fill - expected fill as ratio of original fill.
3091: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3092: Run with the option -info to determine an optimal value to use
3093: .ve
3095: Level: developer
3097: Notes:
3098: Most users should employ the `KSP` interface for linear solvers
3099: instead of working directly with matrix algebra routines such as this.
3100: See, e.g., `KSPCreate()`.
3102: This changes the state of the matrix to a factored matrix; it cannot be used
3103: for example with `MatSetValues()` unless one first calls `MatSetUnfactored()`.
3105: This is really in-place only for dense matrices, the preferred approach is to use `MatGetFactor()`, `MatLUFactorSymbolic()`, and `MatLUFactorNumeric()`
3106: when not using `KSP`.
3108: Developer Note:
3109: The Fortran interface is not autogenerated as the
3110: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3112: .seealso: [](ch_matrices), [Matrix Factorization](sec_matfactor), `Mat`, `MatFactorType`, `MatLUFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`,
3113: `MatGetOrdering()`, `MatSetUnfactored()`, `MatFactorInfo`, `MatGetFactor()`
3114: @*/
3115: PetscErrorCode MatLUFactor(Mat mat, IS row, IS col, const MatFactorInfo *info)
3116: {
3117: MatFactorInfo tinfo;
3119: PetscFunctionBegin;
3123: if (info) PetscAssertPointer(info, 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: MatCheckPreallocated(mat, 1);
3128: if (!info) {
3129: PetscCall(MatFactorInfoInitialize(&tinfo));
3130: info = &tinfo;
3131: }
3133: PetscCall(PetscLogEventBegin(MAT_LUFactor, mat, row, col, 0));
3134: PetscUseTypeMethod(mat, lufactor, row, col, info);
3135: PetscCall(PetscLogEventEnd(MAT_LUFactor, mat, row, col, 0));
3136: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3137: PetscFunctionReturn(PETSC_SUCCESS);
3138: }
3140: /*@
3141: MatILUFactor - Performs in-place ILU factorization of matrix.
3143: Collective
3145: Input Parameters:
3146: + mat - the matrix
3147: . row - row permutation
3148: . col - column permutation
3149: - info - structure containing
3150: .vb
3151: levels - number of levels of fill.
3152: expected fill - as ratio of original fill.
3153: 1 or 0 - indicating force fill on diagonal (improves robustness for matrices
3154: missing diagonal entries)
3155: .ve
3157: Level: developer
3159: Notes:
3160: Most users should employ the `KSP` interface for linear solvers
3161: instead of working directly with matrix algebra routines such as this.
3162: See, e.g., `KSPCreate()`.
3164: Probably really in-place only when level of fill is zero, otherwise allocates
3165: new space to store factored matrix and deletes previous memory. The preferred approach is to use `MatGetFactor()`, `MatILUFactorSymbolic()`, and `MatILUFactorNumeric()`
3166: when not using `KSP`.
3168: Developer Note:
3169: The Fortran interface is not autogenerated as the
3170: interface definition cannot be generated correctly [due to MatFactorInfo]
3172: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatILUFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`, `MatFactorInfo`
3173: @*/
3174: PetscErrorCode MatILUFactor(Mat mat, IS row, IS col, const MatFactorInfo *info)
3175: {
3176: PetscFunctionBegin;
3180: PetscAssertPointer(info, 4);
3182: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "matrix must be square");
3183: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3184: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3185: MatCheckPreallocated(mat, 1);
3187: PetscCall(PetscLogEventBegin(MAT_ILUFactor, mat, row, col, 0));
3188: PetscUseTypeMethod(mat, ilufactor, row, col, info);
3189: PetscCall(PetscLogEventEnd(MAT_ILUFactor, mat, row, col, 0));
3190: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3191: PetscFunctionReturn(PETSC_SUCCESS);
3192: }
3194: /*@
3195: MatLUFactorSymbolic - Performs symbolic LU factorization of matrix.
3196: Call this routine before calling `MatLUFactorNumeric()` and after `MatGetFactor()`.
3198: Collective
3200: Input Parameters:
3201: + fact - the factor matrix obtained with `MatGetFactor()`
3202: . mat - the matrix
3203: . row - the row permutation
3204: . col - the column permutation
3205: - info - options for factorization, includes
3206: .vb
3207: fill - expected fill as ratio of original fill. Run with the option -info to determine an optimal value to use
3208: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3209: .ve
3211: Level: developer
3213: Notes:
3214: See [Matrix Factorization](sec_matfactor) for additional information about factorizations
3216: Most users should employ the simplified `KSP` interface for linear solvers
3217: instead of working directly with matrix algebra routines such as this.
3218: See, e.g., `KSPCreate()`.
3220: Developer Note:
3221: The Fortran interface is not autogenerated as the
3222: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3224: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatLUFactor()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`, `MatFactorInfo`, `MatFactorInfoInitialize()`
3225: @*/
3226: PetscErrorCode MatLUFactorSymbolic(Mat fact, Mat mat, IS row, IS col, const MatFactorInfo *info)
3227: {
3228: MatFactorInfo tinfo;
3230: PetscFunctionBegin;
3235: if (info) PetscAssertPointer(info, 5);
3238: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3239: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3240: MatCheckPreallocated(mat, 2);
3241: if (!info) {
3242: PetscCall(MatFactorInfoInitialize(&tinfo));
3243: info = &tinfo;
3244: }
3246: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_LUFactorSymbolic, mat, row, col, 0));
3247: PetscUseTypeMethod(fact, lufactorsymbolic, mat, row, col, info);
3248: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_LUFactorSymbolic, mat, row, col, 0));
3249: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3250: PetscFunctionReturn(PETSC_SUCCESS);
3251: }
3253: /*@
3254: MatLUFactorNumeric - Performs numeric LU factorization of a matrix.
3255: Call this routine after first calling `MatLUFactorSymbolic()` and `MatGetFactor()`.
3257: Collective
3259: Input Parameters:
3260: + fact - the factor matrix obtained with `MatGetFactor()`
3261: . mat - the matrix
3262: - info - options for factorization
3264: Level: developer
3266: Notes:
3267: See `MatLUFactor()` for in-place factorization. See
3268: `MatCholeskyFactorNumeric()` for the symmetric, positive definite case.
3270: Most users should employ the `KSP` interface for linear solvers
3271: instead of working directly with matrix algebra routines such as this.
3272: See, e.g., `KSPCreate()`.
3274: Developer Note:
3275: The Fortran interface is not autogenerated as the
3276: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3278: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatFactorInfo`, `MatLUFactorSymbolic()`, `MatLUFactor()`, `MatCholeskyFactor()`
3279: @*/
3280: PetscErrorCode MatLUFactorNumeric(Mat fact, Mat mat, const MatFactorInfo *info)
3281: {
3282: MatFactorInfo tinfo;
3284: PetscFunctionBegin;
3289: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3290: 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,
3291: mat->rmap->N, (fact)->rmap->N, mat->cmap->N, (fact)->cmap->N);
3293: MatCheckPreallocated(mat, 2);
3294: if (!info) {
3295: PetscCall(MatFactorInfoInitialize(&tinfo));
3296: info = &tinfo;
3297: }
3299: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_LUFactorNumeric, mat, fact, 0, 0));
3300: else PetscCall(PetscLogEventBegin(MAT_LUFactor, mat, fact, 0, 0));
3301: PetscUseTypeMethod(fact, lufactornumeric, mat, info);
3302: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_LUFactorNumeric, mat, fact, 0, 0));
3303: else PetscCall(PetscLogEventEnd(MAT_LUFactor, mat, fact, 0, 0));
3304: PetscCall(MatViewFromOptions(fact, NULL, "-mat_factor_view"));
3305: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3306: PetscFunctionReturn(PETSC_SUCCESS);
3307: }
3309: /*@
3310: MatCholeskyFactor - Performs in-place Cholesky factorization of a
3311: symmetric matrix.
3313: Collective
3315: Input Parameters:
3316: + mat - the matrix
3317: . perm - row and column permutations
3318: - info - expected fill as ratio of original fill
3320: Level: developer
3322: Notes:
3323: See `MatLUFactor()` for the nonsymmetric case. See also `MatGetFactor()`,
3324: `MatCholeskyFactorSymbolic()`, and `MatCholeskyFactorNumeric()`.
3326: Most users should employ the `KSP` interface for linear solvers
3327: instead of working directly with matrix algebra routines such as this.
3328: See, e.g., `KSPCreate()`.
3330: Developer Note:
3331: The Fortran interface is not autogenerated as the
3332: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3334: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatFactorInfo`, `MatLUFactor()`, `MatCholeskyFactorSymbolic()`, `MatCholeskyFactorNumeric()`
3335: `MatGetOrdering()`
3336: @*/
3337: PetscErrorCode MatCholeskyFactor(Mat mat, IS perm, const MatFactorInfo *info)
3338: {
3339: MatFactorInfo tinfo;
3341: PetscFunctionBegin;
3344: if (info) PetscAssertPointer(info, 3);
3346: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "Matrix must be square");
3347: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3348: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3349: MatCheckPreallocated(mat, 1);
3350: if (!info) {
3351: PetscCall(MatFactorInfoInitialize(&tinfo));
3352: info = &tinfo;
3353: }
3355: PetscCall(PetscLogEventBegin(MAT_CholeskyFactor, mat, perm, 0, 0));
3356: PetscUseTypeMethod(mat, choleskyfactor, perm, info);
3357: PetscCall(PetscLogEventEnd(MAT_CholeskyFactor, mat, perm, 0, 0));
3358: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3359: PetscFunctionReturn(PETSC_SUCCESS);
3360: }
3362: /*@
3363: MatCholeskyFactorSymbolic - Performs symbolic Cholesky factorization
3364: of a symmetric matrix.
3366: Collective
3368: Input Parameters:
3369: + fact - the factor matrix obtained with `MatGetFactor()`
3370: . mat - the matrix
3371: . perm - row and column permutations
3372: - info - options for factorization, includes
3373: .vb
3374: fill - expected fill as ratio of original fill.
3375: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3376: Run with the option -info to determine an optimal value to use
3377: .ve
3379: Level: developer
3381: Notes:
3382: See `MatLUFactorSymbolic()` for the nonsymmetric case. See also
3383: `MatCholeskyFactor()` and `MatCholeskyFactorNumeric()`.
3385: Most users should employ the `KSP` interface for linear solvers
3386: instead of working directly with matrix algebra routines such as this.
3387: See, e.g., `KSPCreate()`.
3389: Developer Note:
3390: The Fortran interface is not autogenerated as the
3391: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3393: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactor()`, `MatCholeskyFactorNumeric()`
3394: `MatGetOrdering()`
3395: @*/
3396: PetscErrorCode MatCholeskyFactorSymbolic(Mat fact, Mat mat, IS perm, const MatFactorInfo *info)
3397: {
3398: MatFactorInfo tinfo;
3400: PetscFunctionBegin;
3404: if (info) PetscAssertPointer(info, 4);
3407: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "Matrix must be square");
3408: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3409: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3410: MatCheckPreallocated(mat, 2);
3411: if (!info) {
3412: PetscCall(MatFactorInfoInitialize(&tinfo));
3413: info = &tinfo;
3414: }
3416: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_CholeskyFactorSymbolic, mat, perm, 0, 0));
3417: PetscUseTypeMethod(fact, choleskyfactorsymbolic, mat, perm, info);
3418: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_CholeskyFactorSymbolic, mat, perm, 0, 0));
3419: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3420: PetscFunctionReturn(PETSC_SUCCESS);
3421: }
3423: /*@
3424: MatCholeskyFactorNumeric - Performs numeric Cholesky factorization
3425: of a symmetric matrix. Call this routine after first calling `MatGetFactor()` and
3426: `MatCholeskyFactorSymbolic()`.
3428: Collective
3430: Input Parameters:
3431: + fact - the factor matrix obtained with `MatGetFactor()`, where the factored values are stored
3432: . mat - the initial matrix that is to be factored
3433: - info - options for factorization
3435: Level: developer
3437: Note:
3438: Most users should employ the `KSP` interface for linear solvers
3439: instead of working directly with matrix algebra routines such as this.
3440: See, e.g., `KSPCreate()`.
3442: Developer Note:
3443: The Fortran interface is not autogenerated as the
3444: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3446: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatCholeskyFactorSymbolic()`, `MatCholeskyFactor()`, `MatLUFactorNumeric()`
3447: @*/
3448: PetscErrorCode MatCholeskyFactorNumeric(Mat fact, Mat mat, const MatFactorInfo *info)
3449: {
3450: MatFactorInfo tinfo;
3452: PetscFunctionBegin;
3457: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3458: 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,
3459: mat->rmap->N, (fact)->rmap->N, mat->cmap->N, (fact)->cmap->N);
3460: MatCheckPreallocated(mat, 2);
3461: if (!info) {
3462: PetscCall(MatFactorInfoInitialize(&tinfo));
3463: info = &tinfo;
3464: }
3466: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_CholeskyFactorNumeric, mat, fact, 0, 0));
3467: else PetscCall(PetscLogEventBegin(MAT_CholeskyFactor, mat, fact, 0, 0));
3468: PetscUseTypeMethod(fact, choleskyfactornumeric, mat, info);
3469: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_CholeskyFactorNumeric, mat, fact, 0, 0));
3470: else PetscCall(PetscLogEventEnd(MAT_CholeskyFactor, mat, fact, 0, 0));
3471: PetscCall(MatViewFromOptions(fact, NULL, "-mat_factor_view"));
3472: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3473: PetscFunctionReturn(PETSC_SUCCESS);
3474: }
3476: /*@
3477: MatQRFactor - Performs in-place QR factorization of matrix.
3479: Collective
3481: Input Parameters:
3482: + mat - the matrix
3483: . col - column permutation
3484: - info - options for factorization, includes
3485: .vb
3486: fill - expected fill as ratio of original fill.
3487: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3488: Run with the option -info to determine an optimal value to use
3489: .ve
3491: Level: developer
3493: Notes:
3494: Most users should employ the `KSP` interface for linear solvers
3495: instead of working directly with matrix algebra routines such as this.
3496: See, e.g., `KSPCreate()`.
3498: This changes the state of the matrix to a factored matrix; it cannot be used
3499: for example with `MatSetValues()` unless one first calls `MatSetUnfactored()`.
3501: Developer Note:
3502: The Fortran interface is not autogenerated as the
3503: interface definition cannot be generated correctly [due to MatFactorInfo]
3505: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatQRFactorSymbolic()`, `MatQRFactorNumeric()`, `MatLUFactor()`,
3506: `MatSetUnfactored()`
3507: @*/
3508: PetscErrorCode MatQRFactor(Mat mat, IS col, const MatFactorInfo *info)
3509: {
3510: PetscFunctionBegin;
3513: if (info) PetscAssertPointer(info, 3);
3515: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3516: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3517: MatCheckPreallocated(mat, 1);
3518: PetscCall(PetscLogEventBegin(MAT_QRFactor, mat, col, 0, 0));
3519: PetscUseMethod(mat, "MatQRFactor_C", (Mat, IS, const MatFactorInfo *), (mat, col, info));
3520: PetscCall(PetscLogEventEnd(MAT_QRFactor, mat, col, 0, 0));
3521: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3522: PetscFunctionReturn(PETSC_SUCCESS);
3523: }
3525: /*@
3526: MatQRFactorSymbolic - Performs symbolic QR factorization of matrix.
3527: Call this routine after `MatGetFactor()` but before calling `MatQRFactorNumeric()`.
3529: Collective
3531: Input Parameters:
3532: + fact - the factor matrix obtained with `MatGetFactor()`
3533: . mat - the matrix
3534: . col - column permutation
3535: - info - options for factorization, includes
3536: .vb
3537: fill - expected fill as ratio of original fill.
3538: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3539: Run with the option -info to determine an optimal value to use
3540: .ve
3542: Level: developer
3544: Note:
3545: Most users should employ the `KSP` interface for linear solvers
3546: instead of working directly with matrix algebra routines such as this.
3547: See, e.g., `KSPCreate()`.
3549: Developer Note:
3550: The Fortran interface is not autogenerated as the
3551: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3553: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatFactorInfo`, `MatQRFactor()`, `MatQRFactorNumeric()`, `MatLUFactor()`, `MatFactorInfoInitialize()`
3554: @*/
3555: PetscErrorCode MatQRFactorSymbolic(Mat fact, Mat mat, IS col, const MatFactorInfo *info)
3556: {
3557: MatFactorInfo tinfo;
3559: PetscFunctionBegin;
3563: if (info) PetscAssertPointer(info, 4);
3566: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3567: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3568: MatCheckPreallocated(mat, 2);
3569: if (!info) {
3570: PetscCall(MatFactorInfoInitialize(&tinfo));
3571: info = &tinfo;
3572: }
3574: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_QRFactorSymbolic, fact, mat, col, 0));
3575: PetscUseMethod(fact, "MatQRFactorSymbolic_C", (Mat, Mat, IS, const MatFactorInfo *), (fact, mat, col, info));
3576: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_QRFactorSymbolic, fact, mat, col, 0));
3577: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3578: PetscFunctionReturn(PETSC_SUCCESS);
3579: }
3581: /*@
3582: MatQRFactorNumeric - Performs numeric QR factorization of a matrix.
3583: Call this routine after first calling `MatGetFactor()`, and `MatQRFactorSymbolic()`.
3585: Collective
3587: Input Parameters:
3588: + fact - the factor matrix obtained with `MatGetFactor()`
3589: . mat - the matrix
3590: - info - options for factorization
3592: Level: developer
3594: Notes:
3595: See `MatQRFactor()` for in-place factorization.
3597: Most users should employ the `KSP` interface for linear solvers
3598: instead of working directly with matrix algebra routines such as this.
3599: See, e.g., `KSPCreate()`.
3601: Developer Note:
3602: The Fortran interface is not autogenerated as the
3603: interface definition cannot be generated correctly [due to `MatFactorInfo`]
3605: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatQRFactor()`, `MatQRFactorSymbolic()`, `MatLUFactor()`
3606: @*/
3607: PetscErrorCode MatQRFactorNumeric(Mat fact, Mat mat, const MatFactorInfo *info)
3608: {
3609: MatFactorInfo tinfo;
3611: PetscFunctionBegin;
3616: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3617: 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,
3618: mat->rmap->N, (fact)->rmap->N, mat->cmap->N, (fact)->cmap->N);
3620: MatCheckPreallocated(mat, 2);
3621: if (!info) {
3622: PetscCall(MatFactorInfoInitialize(&tinfo));
3623: info = &tinfo;
3624: }
3626: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_QRFactorNumeric, mat, fact, 0, 0));
3627: else PetscCall(PetscLogEventBegin(MAT_QRFactor, mat, fact, 0, 0));
3628: PetscUseMethod(fact, "MatQRFactorNumeric_C", (Mat, Mat, const MatFactorInfo *), (fact, mat, info));
3629: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_QRFactorNumeric, mat, fact, 0, 0));
3630: else PetscCall(PetscLogEventEnd(MAT_QRFactor, mat, fact, 0, 0));
3631: PetscCall(MatViewFromOptions(fact, NULL, "-mat_factor_view"));
3632: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3633: PetscFunctionReturn(PETSC_SUCCESS);
3634: }
3636: /*@
3637: MatSolve - Solves $A x = b$, given a factored matrix.
3639: Neighbor-wise Collective
3641: Input Parameters:
3642: + mat - the factored matrix
3643: - b - the right-hand-side vector
3645: Output Parameter:
3646: . x - the result vector
3648: Level: developer
3650: Notes:
3651: The vectors `b` and `x` cannot be the same. I.e., one cannot
3652: call `MatSolve`(A,x,x).
3654: Most users should employ the `KSP` interface for linear solvers
3655: instead of working directly with matrix algebra routines such as this.
3656: See, e.g., `KSPCreate()`.
3658: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatLUFactor()`, `MatSolveAdd()`, `MatSolveTranspose()`, `MatSolveTransposeAdd()`
3659: @*/
3660: PetscErrorCode MatSolve(Mat mat, Vec b, Vec x)
3661: {
3662: PetscFunctionBegin;
3667: PetscCheckSameComm(mat, 1, b, 2);
3668: PetscCheckSameComm(mat, 1, x, 3);
3669: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
3670: 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);
3671: 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);
3672: 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);
3673: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3674: MatCheckPreallocated(mat, 1);
3676: PetscCall(PetscLogEventBegin(MAT_Solve, mat, b, x, 0));
3677: PetscCall(VecFlag(x, mat->factorerrortype));
3678: if (mat->factorerrortype) {
3679: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
3680: } else PetscUseTypeMethod(mat, solve, b, x);
3681: PetscCall(PetscLogEventEnd(MAT_Solve, mat, b, x, 0));
3682: PetscCall(PetscObjectStateIncrease((PetscObject)x));
3683: PetscFunctionReturn(PETSC_SUCCESS);
3684: }
3686: static PetscErrorCode MatMatSolve_Basic(Mat A, Mat B, Mat X, PetscBool trans)
3687: {
3688: Vec b, x;
3689: PetscInt N, i;
3690: PetscErrorCode (*f)(Mat, Vec, Vec);
3691: PetscBool Abound, Bneedconv = PETSC_FALSE, Xneedconv = PETSC_FALSE;
3693: PetscFunctionBegin;
3694: if (A->factorerrortype) {
3695: PetscCall(PetscInfo(A, "MatFactorError %d\n", A->factorerrortype));
3696: PetscCall(MatSetInf(X));
3697: PetscFunctionReturn(PETSC_SUCCESS);
3698: }
3699: f = (!trans || (!A->ops->solvetranspose && A->symmetric)) ? A->ops->solve : A->ops->solvetranspose;
3700: PetscCheck(f, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Mat type %s", ((PetscObject)A)->type_name);
3701: PetscCall(MatBoundToCPU(A, &Abound));
3702: if (!Abound) {
3703: PetscCall(PetscObjectTypeCompareAny((PetscObject)B, &Bneedconv, MATSEQDENSE, MATMPIDENSE, ""));
3704: PetscCall(PetscObjectTypeCompareAny((PetscObject)X, &Xneedconv, MATSEQDENSE, MATMPIDENSE, ""));
3705: }
3706: #if PetscDefined(HAVE_CUDA)
3707: if (Bneedconv) PetscCall(MatConvert(B, MATDENSECUDA, MAT_INPLACE_MATRIX, &B));
3708: if (Xneedconv) PetscCall(MatConvert(X, MATDENSECUDA, MAT_INPLACE_MATRIX, &X));
3709: #elif PetscDefined(HAVE_HIP)
3710: if (Bneedconv) PetscCall(MatConvert(B, MATDENSEHIP, MAT_INPLACE_MATRIX, &B));
3711: if (Xneedconv) PetscCall(MatConvert(X, MATDENSEHIP, MAT_INPLACE_MATRIX, &X));
3712: #endif
3713: PetscCall(MatGetSize(B, NULL, &N));
3714: for (i = 0; i < N; i++) {
3715: PetscCall(MatDenseGetColumnVecRead(B, i, &b));
3716: PetscCall(MatDenseGetColumnVecWrite(X, i, &x));
3717: PetscCall((*f)(A, b, x));
3718: PetscCall(MatDenseRestoreColumnVecWrite(X, i, &x));
3719: PetscCall(MatDenseRestoreColumnVecRead(B, i, &b));
3720: }
3721: if (Bneedconv) PetscCall(MatConvert(B, MATDENSE, MAT_INPLACE_MATRIX, &B));
3722: if (Xneedconv) PetscCall(MatConvert(X, MATDENSE, MAT_INPLACE_MATRIX, &X));
3723: PetscFunctionReturn(PETSC_SUCCESS);
3724: }
3726: /*@
3727: MatMatSolve - Solves $A X = B$, given a factored matrix.
3729: Neighbor-wise Collective
3731: Input Parameters:
3732: + A - the factored matrix
3733: - B - the right-hand-side matrix `MATDENSE` (or sparse `MATAIJ`-- when using MUMPS)
3735: Output Parameter:
3736: . X - the result matrix (dense matrix)
3738: Level: developer
3740: Note:
3741: If `B` is a `MATDENSE` matrix then one can call `MatMatSolve`(A,B,B) except with `MATSOLVERMKL_CPARDISO`;
3742: otherwise, `B` and `X` cannot be the same.
3744: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSolve()`, `MatMatSolveTranspose()`, `MatLUFactor()`, `MatCholeskyFactor()`
3745: @*/
3746: PetscErrorCode MatMatSolve(Mat A, Mat B, Mat X)
3747: {
3748: PetscFunctionBegin;
3753: PetscCheckSameComm(A, 1, B, 2);
3754: PetscCheckSameComm(A, 1, X, 3);
3755: 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);
3756: 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);
3757: 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");
3758: if (!A->rmap->N && !A->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3759: MatCheckPreallocated(A, 1);
3761: PetscCall(PetscLogEventBegin(MAT_MatSolve, A, B, X, 0));
3762: if (!A->ops->matsolve) {
3763: PetscCall(PetscInfo(A, "Mat type %s using basic MatMatSolve\n", ((PetscObject)A)->type_name));
3764: PetscCall(MatMatSolve_Basic(A, B, X, PETSC_FALSE));
3765: } else PetscUseTypeMethod(A, matsolve, B, X);
3766: PetscCall(PetscLogEventEnd(MAT_MatSolve, A, B, X, 0));
3767: PetscCall(PetscObjectStateIncrease((PetscObject)X));
3768: PetscFunctionReturn(PETSC_SUCCESS);
3769: }
3771: /*@
3772: MatMatSolveTranspose - Solves $A^T X = B $, given a factored matrix.
3774: Neighbor-wise Collective
3776: Input Parameters:
3777: + A - the factored matrix
3778: - B - the right-hand-side matrix (`MATDENSE` matrix)
3780: Output Parameter:
3781: . X - the result matrix (dense matrix)
3783: Level: developer
3785: Note:
3786: The matrices `B` and `X` cannot be the same. I.e., one cannot
3787: call `MatMatSolveTranspose`(A,X,X).
3789: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSolveTranspose()`, `MatMatSolve()`, `MatLUFactor()`, `MatCholeskyFactor()`
3790: @*/
3791: PetscErrorCode MatMatSolveTranspose(Mat A, Mat B, Mat X)
3792: {
3793: PetscFunctionBegin;
3798: PetscCheckSameComm(A, 1, B, 2);
3799: PetscCheckSameComm(A, 1, X, 3);
3800: PetscCheck(X != B, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_IDN, "X and B must be different matrices");
3801: 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);
3802: 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);
3803: 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);
3804: 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");
3805: if (!A->rmap->N && !A->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3806: MatCheckPreallocated(A, 1);
3808: PetscCall(PetscLogEventBegin(MAT_MatSolve, A, B, X, 0));
3809: if (!A->ops->matsolvetranspose) {
3810: PetscCall(PetscInfo(A, "Mat type %s using basic MatMatSolveTranspose\n", ((PetscObject)A)->type_name));
3811: PetscCall(MatMatSolve_Basic(A, B, X, PETSC_TRUE));
3812: } else PetscUseTypeMethod(A, matsolvetranspose, B, X);
3813: PetscCall(PetscLogEventEnd(MAT_MatSolve, A, B, X, 0));
3814: PetscCall(PetscObjectStateIncrease((PetscObject)X));
3815: PetscFunctionReturn(PETSC_SUCCESS);
3816: }
3818: /*@
3819: MatMatTransposeSolve - Solves $A X = B^T$, given a factored matrix.
3821: Neighbor-wise Collective
3823: Input Parameters:
3824: + A - the factored matrix
3825: - Bt - the transpose of right-hand-side matrix as a `MATDENSE`
3827: Output Parameter:
3828: . X - the result matrix (dense matrix)
3830: Level: developer
3832: Note:
3833: For MUMPS, it only supports centralized sparse compressed column format on the host processor for right-hand side matrix. User must create `Bt` in sparse compressed row
3834: format on the host processor and call `MatMatTransposeSolve()` to implement MUMPS' `MatMatSolve()`.
3836: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatMatSolve()`, `MatMatSolveTranspose()`, `MatLUFactor()`, `MatCholeskyFactor()`
3837: @*/
3838: PetscErrorCode MatMatTransposeSolve(Mat A, Mat Bt, Mat X)
3839: {
3840: PetscFunctionBegin;
3845: PetscCheckSameComm(A, 1, Bt, 2);
3846: PetscCheckSameComm(A, 1, X, 3);
3848: PetscCheck(X != Bt, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_IDN, "X and B must be different matrices");
3849: 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);
3850: 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);
3851: 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");
3852: if (!A->rmap->N && !A->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3853: PetscCheck(A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Unfactored matrix");
3854: MatCheckPreallocated(A, 1);
3856: PetscCall(PetscLogEventBegin(MAT_MatTrSolve, A, Bt, X, 0));
3857: PetscUseTypeMethod(A, mattransposesolve, Bt, X);
3858: PetscCall(PetscLogEventEnd(MAT_MatTrSolve, A, Bt, X, 0));
3859: PetscCall(PetscObjectStateIncrease((PetscObject)X));
3860: PetscFunctionReturn(PETSC_SUCCESS);
3861: }
3863: /*@
3864: MatForwardSolve - Solves $ L x = b $, given a factored matrix, $A = LU $, or
3865: $U^T*D^(1/2) x = b$, given a factored symmetric matrix, $A = U^T*D*U$,
3867: Neighbor-wise Collective
3869: Input Parameters:
3870: + mat - the factored matrix
3871: - b - the right-hand-side vector
3873: Output Parameter:
3874: . x - the result vector
3876: Level: developer
3878: Notes:
3879: `MatSolve()` should be used for most applications, as it performs
3880: a forward solve followed by a backward solve.
3882: The vectors `b` and `x` cannot be the same, i.e., one cannot
3883: call `MatForwardSolve`(A,x,x).
3885: For matrix in `MATSEQBAIJ` format with block size larger than 1,
3886: the diagonal blocks are not implemented as $D = D^(1/2) * D^(1/2)$ yet.
3887: `MatForwardSolve()` solves $U^T*D y = b$, and
3888: `MatBackwardSolve()` solves $U x = y$.
3889: Thus they do not provide a symmetric preconditioner.
3891: .seealso: [](ch_matrices), `Mat`, `MatBackwardSolve()`, `MatGetFactor()`, `MatSolve()`
3892: @*/
3893: PetscErrorCode MatForwardSolve(Mat mat, Vec b, Vec x)
3894: {
3895: PetscFunctionBegin;
3900: PetscCheckSameComm(mat, 1, b, 2);
3901: PetscCheckSameComm(mat, 1, x, 3);
3902: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
3903: 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);
3904: 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);
3905: 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);
3906: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3907: MatCheckPreallocated(mat, 1);
3909: PetscCall(PetscLogEventBegin(MAT_ForwardSolve, mat, b, x, 0));
3910: PetscUseTypeMethod(mat, forwardsolve, b, x);
3911: PetscCall(PetscLogEventEnd(MAT_ForwardSolve, mat, b, x, 0));
3912: PetscCall(PetscObjectStateIncrease((PetscObject)x));
3913: PetscFunctionReturn(PETSC_SUCCESS);
3914: }
3916: /*@
3917: MatBackwardSolve - Solves $U x = b$, given a factored matrix, $A = LU$.
3918: $D^(1/2) U x = b$, given a factored symmetric matrix, $A = U^T*D*U$,
3920: Neighbor-wise Collective
3922: Input Parameters:
3923: + mat - the factored matrix
3924: - b - the right-hand-side vector
3926: Output Parameter:
3927: . x - the result vector
3929: Level: developer
3931: Notes:
3932: `MatSolve()` should be used for most applications, as it performs
3933: a forward solve followed by a backward solve.
3935: The vectors `b` and `x` cannot be the same. I.e., one cannot
3936: call `MatBackwardSolve`(A,x,x).
3938: For matrix in `MATSEQBAIJ` format with block size larger than 1,
3939: the diagonal blocks are not implemented as $D = D^(1/2) * D^(1/2)$ yet.
3940: `MatForwardSolve()` solves $U^T*D y = b$, and
3941: `MatBackwardSolve()` solves $U x = y$.
3942: Thus they do not provide a symmetric preconditioner.
3944: .seealso: [](ch_matrices), `Mat`, `MatForwardSolve()`, `MatGetFactor()`, `MatSolve()`
3945: @*/
3946: PetscErrorCode MatBackwardSolve(Mat mat, Vec b, Vec x)
3947: {
3948: PetscFunctionBegin;
3953: PetscCheckSameComm(mat, 1, b, 2);
3954: PetscCheckSameComm(mat, 1, x, 3);
3955: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
3956: 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);
3957: 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);
3958: 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);
3959: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3960: MatCheckPreallocated(mat, 1);
3962: PetscCall(PetscLogEventBegin(MAT_BackwardSolve, mat, b, x, 0));
3963: PetscUseTypeMethod(mat, backwardsolve, b, x);
3964: PetscCall(PetscLogEventEnd(MAT_BackwardSolve, mat, b, x, 0));
3965: PetscCall(PetscObjectStateIncrease((PetscObject)x));
3966: PetscFunctionReturn(PETSC_SUCCESS);
3967: }
3969: /*@
3970: MatSolveAdd - Computes $x = y + A^{-1}*b$, given a factored matrix.
3972: Neighbor-wise Collective
3974: Input Parameters:
3975: + mat - the factored matrix
3976: . b - the right-hand-side vector
3977: - y - the vector to be added to
3979: Output Parameter:
3980: . x - the result vector
3982: Level: developer
3984: Note:
3985: The vectors `b` and `x` cannot be the same. I.e., one cannot
3986: call `MatSolveAdd`(A,x,y,x).
3988: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatSolve()`, `MatGetFactor()`, `MatSolveTranspose()`, `MatSolveTransposeAdd()`
3989: @*/
3990: PetscErrorCode MatSolveAdd(Mat mat, Vec b, Vec y, Vec x)
3991: {
3992: PetscScalar one = 1.0;
3993: Vec tmp;
3995: PetscFunctionBegin;
4001: PetscCheckSameComm(mat, 1, b, 2);
4002: PetscCheckSameComm(mat, 1, y, 3);
4003: PetscCheckSameComm(mat, 1, x, 4);
4004: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4005: 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);
4006: 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);
4007: 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);
4008: 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);
4009: 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);
4010: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4011: MatCheckPreallocated(mat, 1);
4013: PetscCall(PetscLogEventBegin(MAT_SolveAdd, mat, b, x, y));
4014: PetscCall(VecFlag(x, mat->factorerrortype));
4015: if (mat->factorerrortype) {
4016: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
4017: } else if (mat->ops->solveadd) {
4018: PetscUseTypeMethod(mat, solveadd, b, y, x);
4019: } else {
4020: /* do the solve then the add manually */
4021: if (x != y) {
4022: PetscCall(MatSolve(mat, b, x));
4023: PetscCall(VecAXPY(x, one, y));
4024: } else {
4025: PetscCall(VecDuplicate(x, &tmp));
4026: PetscCall(VecCopy(x, tmp));
4027: PetscCall(MatSolve(mat, b, x));
4028: PetscCall(VecAXPY(x, one, tmp));
4029: PetscCall(VecDestroy(&tmp));
4030: }
4031: }
4032: PetscCall(PetscLogEventEnd(MAT_SolveAdd, mat, b, x, y));
4033: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4034: PetscFunctionReturn(PETSC_SUCCESS);
4035: }
4037: /*@
4038: MatSolveTranspose - Solves $A^T x = b$, given a factored matrix.
4040: Neighbor-wise Collective
4042: Input Parameters:
4043: + mat - the factored matrix
4044: - b - the right-hand-side vector
4046: Output Parameter:
4047: . x - the result vector
4049: Level: developer
4051: Notes:
4052: The vectors `b` and `x` cannot be the same. I.e., one cannot
4053: call `MatSolveTranspose`(A,x,x).
4055: Most users should employ the `KSP` interface for linear solvers
4056: instead of working directly with matrix algebra routines such as this.
4057: See, e.g., `KSPCreate()`.
4059: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `KSP`, `MatSolve()`, `MatSolveAdd()`, `MatSolveTransposeAdd()`
4060: @*/
4061: PetscErrorCode MatSolveTranspose(Mat mat, Vec b, Vec x)
4062: {
4063: PetscErrorCode (*f)(Mat, Vec, Vec) = (!mat->ops->solvetranspose && mat->symmetric) ? mat->ops->solve : mat->ops->solvetranspose;
4065: PetscFunctionBegin;
4070: PetscCheckSameComm(mat, 1, b, 2);
4071: PetscCheckSameComm(mat, 1, x, 3);
4072: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4073: 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);
4074: 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);
4075: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4076: MatCheckPreallocated(mat, 1);
4077: PetscCall(PetscLogEventBegin(MAT_SolveTranspose, mat, b, x, 0));
4078: PetscCall(VecFlag(x, mat->factorerrortype));
4079: if (mat->factorerrortype) {
4080: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
4081: } else {
4082: PetscCheck(f, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Matrix type %s", ((PetscObject)mat)->type_name);
4083: PetscCall((*f)(mat, b, x));
4084: }
4085: PetscCall(PetscLogEventEnd(MAT_SolveTranspose, mat, b, x, 0));
4086: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4087: PetscFunctionReturn(PETSC_SUCCESS);
4088: }
4090: /*@
4091: MatSolveTransposeAdd - Computes $x = y + A^{-T} b$
4092: factored matrix.
4094: Neighbor-wise Collective
4096: Input Parameters:
4097: + mat - the factored matrix
4098: . b - the right-hand-side vector
4099: - y - the vector to be added to
4101: Output Parameter:
4102: . x - the result vector
4104: Level: developer
4106: Note:
4107: The vectors `b` and `x` cannot be the same. I.e., one cannot
4108: call `MatSolveTransposeAdd`(A,x,y,x).
4110: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatSolve()`, `MatSolveAdd()`, `MatSolveTranspose()`
4111: @*/
4112: PetscErrorCode MatSolveTransposeAdd(Mat mat, Vec b, Vec y, Vec x)
4113: {
4114: PetscScalar one = 1.0;
4115: Vec tmp;
4116: PetscErrorCode (*f)(Mat, Vec, Vec, Vec) = (!mat->ops->solvetransposeadd && mat->symmetric) ? mat->ops->solveadd : mat->ops->solvetransposeadd;
4118: PetscFunctionBegin;
4124: PetscCheckSameComm(mat, 1, b, 2);
4125: PetscCheckSameComm(mat, 1, y, 3);
4126: PetscCheckSameComm(mat, 1, x, 4);
4127: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4128: 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);
4129: 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);
4130: 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);
4131: 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);
4132: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4133: MatCheckPreallocated(mat, 1);
4135: PetscCall(PetscLogEventBegin(MAT_SolveTransposeAdd, mat, b, x, y));
4136: PetscCall(VecFlag(x, mat->factorerrortype));
4137: if (mat->factorerrortype) {
4138: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
4139: } else if (f) {
4140: PetscCall((*f)(mat, b, y, x));
4141: } else {
4142: /* do the solve then the add manually */
4143: if (x != y) {
4144: PetscCall(MatSolveTranspose(mat, b, x));
4145: PetscCall(VecAXPY(x, one, y));
4146: } else {
4147: PetscCall(VecDuplicate(x, &tmp));
4148: PetscCall(VecCopy(x, tmp));
4149: PetscCall(MatSolveTranspose(mat, b, x));
4150: PetscCall(VecAXPY(x, one, tmp));
4151: PetscCall(VecDestroy(&tmp));
4152: }
4153: }
4154: PetscCall(PetscLogEventEnd(MAT_SolveTransposeAdd, mat, b, x, y));
4155: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4156: PetscFunctionReturn(PETSC_SUCCESS);
4157: }
4159: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
4160: /*@
4161: MatSOR - Computes relaxation (SOR, Gauss-Seidel) sweeps.
4163: Neighbor-wise Collective
4165: Input Parameters:
4166: + mat - the matrix
4167: . b - the right-hand side
4168: . omega - the relaxation factor
4169: . flag - flag indicating the type of SOR (see below)
4170: . shift - diagonal shift
4171: . its - the number of iterations
4172: - lits - the number of local iterations
4174: Output Parameter:
4175: . x - the solution (can contain an initial guess, use option `SOR_ZERO_INITIAL_GUESS` to indicate no guess)
4177: SOR Flags:
4178: + `SOR_FORWARD_SWEEP` - forward SOR
4179: . `SOR_BACKWARD_SWEEP` - backward SOR
4180: . `SOR_SYMMETRIC_SWEEP` - SSOR (symmetric SOR)
4181: . `SOR_LOCAL_FORWARD_SWEEP` - local forward SOR
4182: . `SOR_LOCAL_BACKWARD_SWEEP` - local forward SOR
4183: . `SOR_LOCAL_SYMMETRIC_SWEEP` - local SSOR
4184: . `SOR_EISENSTAT` - SOR with Eisenstat trick
4185: . `SOR_APPLY_UPPER`, `SOR_APPLY_LOWER` - applies
4186: upper/lower triangular part of matrix to
4187: vector (with omega)
4188: - `SOR_ZERO_INITIAL_GUESS` - zero initial guess
4190: Level: developer
4192: Notes:
4193: `SOR_LOCAL_FORWARD_SWEEP`, `SOR_LOCAL_BACKWARD_SWEEP`, and
4194: `SOR_LOCAL_SYMMETRIC_SWEEP` perform separate independent smoothings
4195: on each processor.
4197: Application programmers will not generally use `MatSOR()` directly,
4198: but instead will employ the `KSP`/`PC` interface.
4200: For `MATBAIJ`, `MATSBAIJ`, and `MATAIJ` matrices with Inodes this does a block SOR smoothing, otherwise it does a pointwise smoothing
4202: Most users should employ the `KSP` interface for linear solvers
4203: instead of working directly with matrix algebra routines such as this.
4204: See, e.g., `KSPCreate()`.
4206: Vectors `x` and `b` CANNOT be the same
4208: The flags are implemented as bitwise inclusive or operations.
4209: For example, use (`SOR_ZERO_INITIAL_GUESS` | `SOR_SYMMETRIC_SWEEP`)
4210: to specify a zero initial guess for SSOR.
4212: Developer Note:
4213: We should add block SOR support for `MATAIJ` matrices with block size set to great than one and no inodes
4215: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `KSP`, `PC`, `MatGetFactor()`
4216: @*/
4217: PetscErrorCode MatSOR(Mat mat, Vec b, PetscReal omega, MatSORType flag, PetscReal shift, PetscInt its, PetscInt lits, Vec x)
4218: {
4219: PetscFunctionBegin;
4224: PetscCheckSameComm(mat, 1, b, 2);
4225: PetscCheckSameComm(mat, 1, x, 8);
4226: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
4227: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4228: 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);
4229: 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);
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(its > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Relaxation requires global its %" PetscInt_FMT " positive", its);
4232: PetscCheck(lits > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Relaxation requires local its %" PetscInt_FMT " positive", lits);
4233: PetscCheck(b != x, PETSC_COMM_SELF, PETSC_ERR_ARG_IDN, "b and x vector cannot be the same");
4235: MatCheckPreallocated(mat, 1);
4236: PetscCall(PetscLogEventBegin(MAT_SOR, mat, b, x, 0));
4237: PetscUseTypeMethod(mat, sor, b, omega, flag, shift, its, lits, x);
4238: PetscCall(PetscLogEventEnd(MAT_SOR, mat, b, x, 0));
4239: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4240: PetscFunctionReturn(PETSC_SUCCESS);
4241: }
4243: /*
4244: Default matrix copy routine.
4245: */
4246: PetscErrorCode MatCopy_Basic(Mat A, Mat B, MatStructure str)
4247: {
4248: PetscInt i, rstart = 0, rend = 0, nz;
4249: const PetscInt *cwork;
4250: const PetscScalar *vwork;
4252: PetscFunctionBegin;
4253: if (B->assembled) PetscCall(MatZeroEntries(B));
4254: if (str == SAME_NONZERO_PATTERN) {
4255: PetscCall(MatGetOwnershipRange(A, &rstart, &rend));
4256: for (i = rstart; i < rend; i++) {
4257: PetscCall(MatGetRow(A, i, &nz, &cwork, &vwork));
4258: PetscCall(MatSetValues(B, 1, &i, nz, cwork, vwork, INSERT_VALUES));
4259: PetscCall(MatRestoreRow(A, i, &nz, &cwork, &vwork));
4260: }
4261: } else {
4262: PetscCall(MatAYPX(B, 0.0, A, str));
4263: }
4264: PetscCall(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY));
4265: PetscCall(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY));
4266: PetscFunctionReturn(PETSC_SUCCESS);
4267: }
4269: /*@
4270: MatCopy - Copies a matrix to another matrix.
4272: Collective
4274: Input Parameters:
4275: + A - the matrix
4276: - str - `SAME_NONZERO_PATTERN` or `DIFFERENT_NONZERO_PATTERN`
4278: Output Parameter:
4279: . B - where the copy is put
4281: Level: intermediate
4283: Notes:
4284: If you use `SAME_NONZERO_PATTERN`, then the two matrices must have the same nonzero pattern or the routine will crash.
4286: `MatCopy()` copies the matrix entries of a matrix to another existing
4287: matrix (after first zeroing the second matrix). A related routine is
4288: `MatConvert()`, which first creates a new matrix and then copies the data.
4290: .seealso: [](ch_matrices), `Mat`, `MatConvert()`, `MatDuplicate()`
4291: @*/
4292: PetscErrorCode MatCopy(Mat A, Mat B, MatStructure str)
4293: {
4294: PetscInt i;
4296: PetscFunctionBegin;
4301: PetscCheckSameComm(A, 1, B, 2);
4302: MatCheckPreallocated(B, 2);
4303: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
4304: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4305: 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,
4306: A->cmap->N, B->cmap->N);
4307: MatCheckPreallocated(A, 1);
4308: if (A == B) PetscFunctionReturn(PETSC_SUCCESS);
4310: PetscCall(PetscLogEventBegin(MAT_Copy, A, B, 0, 0));
4311: if (A->ops->copy) PetscUseTypeMethod(A, copy, B, str);
4312: else PetscCall(MatCopy_Basic(A, B, str));
4314: B->stencil.dim = A->stencil.dim;
4315: B->stencil.noc = A->stencil.noc;
4316: for (i = 0; i <= A->stencil.dim + (A->stencil.noc ? 0 : -1); i++) {
4317: B->stencil.dims[i] = A->stencil.dims[i];
4318: B->stencil.starts[i] = A->stencil.starts[i];
4319: }
4321: PetscCall(PetscLogEventEnd(MAT_Copy, A, B, 0, 0));
4322: PetscCall(PetscObjectStateIncrease((PetscObject)B));
4323: PetscFunctionReturn(PETSC_SUCCESS);
4324: }
4326: /*@
4327: MatConvert - Converts a matrix to another matrix, either of the same
4328: or different type.
4330: Collective
4332: Input Parameters:
4333: + mat - the matrix
4334: . newtype - new matrix type. Use `MATSAME` to create a new matrix of the
4335: same type as the original matrix.
4336: - reuse - denotes if the destination matrix is to be created or reused.
4337: 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
4338: `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).
4340: Output Parameter:
4341: . M - pointer to place new matrix
4343: Level: intermediate
4345: Notes:
4346: `MatConvert()` first creates a new matrix and then copies the data from
4347: the first matrix. A related routine is `MatCopy()`, which copies the matrix
4348: entries of one matrix to another already existing matrix context.
4350: Cannot be used to convert a sequential matrix to parallel or parallel to sequential,
4351: the MPI communicator of the generated matrix is always the same as the communicator
4352: of the input matrix.
4354: .seealso: [](ch_matrices), `Mat`, `MatCopy()`, `MatDuplicate()`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`
4355: @*/
4356: PetscErrorCode MatConvert(Mat mat, MatType newtype, MatReuse reuse, Mat *M)
4357: {
4358: PetscBool sametype, issame, flg;
4359: PetscBool3 issymmetric, ishermitian;
4360: char convname[256], mtype[256];
4361: Mat B;
4363: PetscFunctionBegin;
4366: PetscAssertPointer(M, 4);
4367: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
4368: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4369: MatCheckPreallocated(mat, 1);
4371: PetscCall(PetscOptionsGetString(((PetscObject)mat)->options, ((PetscObject)mat)->prefix, "-matconvert_type", mtype, sizeof(mtype), &flg));
4372: if (flg) newtype = mtype;
4374: PetscCall(PetscObjectTypeCompare((PetscObject)mat, newtype, &sametype));
4375: PetscCall(PetscStrcmp(newtype, "same", &issame));
4376: PetscCheck(!(reuse == MAT_INPLACE_MATRIX) || !(mat != *M), PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "MAT_INPLACE_MATRIX requires same input and output matrix");
4377: if (reuse == MAT_REUSE_MATRIX) {
4379: PetscCheck(mat != *M, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "MAT_REUSE_MATRIX means reuse matrix in final argument, perhaps you mean MAT_INPLACE_MATRIX");
4380: }
4382: if ((reuse == MAT_INPLACE_MATRIX) && (issame || sametype)) {
4383: PetscCall(PetscInfo(mat, "Early return for inplace %s %d %d\n", ((PetscObject)mat)->type_name, sametype, issame));
4384: PetscFunctionReturn(PETSC_SUCCESS);
4385: }
4387: /* Cache Mat options because some converters use MatHeaderReplace */
4388: issymmetric = mat->symmetric;
4389: ishermitian = mat->hermitian;
4391: if ((sametype || issame) && (reuse == MAT_INITIAL_MATRIX) && mat->ops->duplicate) {
4392: PetscCall(PetscInfo(mat, "Calling duplicate for initial matrix %s %d %d\n", ((PetscObject)mat)->type_name, sametype, issame));
4393: PetscUseTypeMethod(mat, duplicate, MAT_COPY_VALUES, M);
4394: } else {
4395: PetscErrorCode (*conv)(Mat, MatType, MatReuse, Mat *) = NULL;
4396: const char *prefix[3] = {"seq", "mpi", ""};
4397: PetscInt i;
4398: /*
4399: Order of precedence:
4400: 0) See if newtype is a superclass of the current matrix.
4401: 1) See if a specialized converter is known to the current matrix.
4402: 2) See if a specialized converter is known to the desired matrix class.
4403: 3) See if a good general converter is registered for the desired class
4404: (as of 6/27/03 only MATMPIADJ falls into this category).
4405: 4) See if a good general converter is known for the current matrix.
4406: 5) Use a really basic converter.
4407: */
4409: /* 0) See if newtype is a superclass of the current matrix.
4410: i.e mat is mpiaij and newtype is aij */
4411: for (i = 0; i < 2; i++) {
4412: PetscCall(PetscStrncpy(convname, prefix[i], sizeof(convname)));
4413: PetscCall(PetscStrlcat(convname, newtype, sizeof(convname)));
4414: PetscCall(PetscStrcmp(convname, ((PetscObject)mat)->type_name, &flg));
4415: PetscCall(PetscInfo(mat, "Check superclass %s %s -> %d\n", convname, ((PetscObject)mat)->type_name, flg));
4416: if (flg) {
4417: if (reuse == MAT_INPLACE_MATRIX) {
4418: PetscCall(PetscInfo(mat, "Early return\n"));
4419: PetscFunctionReturn(PETSC_SUCCESS);
4420: } else if (reuse == MAT_INITIAL_MATRIX && mat->ops->duplicate) {
4421: PetscCall(PetscInfo(mat, "Calling MatDuplicate\n"));
4422: PetscUseTypeMethod(mat, duplicate, MAT_COPY_VALUES, M);
4423: PetscFunctionReturn(PETSC_SUCCESS);
4424: } else if (reuse == MAT_REUSE_MATRIX && mat->ops->copy) {
4425: PetscCall(PetscInfo(mat, "Calling MatCopy\n"));
4426: PetscCall(MatCopy(mat, *M, SAME_NONZERO_PATTERN));
4427: PetscFunctionReturn(PETSC_SUCCESS);
4428: }
4429: }
4430: }
4431: /* 1) See if a specialized converter is known to the current matrix and the desired class */
4432: for (i = 0; i < 3; i++) {
4433: PetscCall(PetscStrncpy(convname, "MatConvert_", sizeof(convname)));
4434: PetscCall(PetscStrlcat(convname, ((PetscObject)mat)->type_name, sizeof(convname)));
4435: PetscCall(PetscStrlcat(convname, "_", sizeof(convname)));
4436: PetscCall(PetscStrlcat(convname, prefix[i], sizeof(convname)));
4437: PetscCall(PetscStrlcat(convname, issame ? ((PetscObject)mat)->type_name : newtype, sizeof(convname)));
4438: PetscCall(PetscStrlcat(convname, "_C", sizeof(convname)));
4439: PetscCall(PetscObjectQueryFunction((PetscObject)mat, convname, &conv));
4440: PetscCall(PetscInfo(mat, "Check specialized (1) %s (%s) -> %d\n", convname, ((PetscObject)mat)->type_name, !!conv));
4441: if (conv) goto foundconv;
4442: }
4444: /* 2) See if a specialized converter is known to the desired matrix class. */
4445: PetscCall(MatCreate(PetscObjectComm((PetscObject)mat), &B));
4446: PetscCall(MatSetSizes(B, mat->rmap->n, mat->cmap->n, mat->rmap->N, mat->cmap->N));
4447: PetscCall(MatSetType(B, newtype));
4448: for (i = 0; i < 3; i++) {
4449: PetscCall(PetscStrncpy(convname, "MatConvert_", sizeof(convname)));
4450: PetscCall(PetscStrlcat(convname, ((PetscObject)mat)->type_name, sizeof(convname)));
4451: PetscCall(PetscStrlcat(convname, "_", sizeof(convname)));
4452: PetscCall(PetscStrlcat(convname, prefix[i], sizeof(convname)));
4453: PetscCall(PetscStrlcat(convname, newtype, sizeof(convname)));
4454: PetscCall(PetscStrlcat(convname, "_C", sizeof(convname)));
4455: PetscCall(PetscObjectQueryFunction((PetscObject)B, convname, &conv));
4456: PetscCall(PetscInfo(mat, "Check specialized (2) %s (%s) -> %d\n", convname, ((PetscObject)B)->type_name, !!conv));
4457: if (conv) {
4458: PetscCall(MatDestroy(&B));
4459: goto foundconv;
4460: }
4461: }
4463: /* 3) See if a good general converter is registered for the desired class */
4464: conv = B->ops->convertfrom;
4465: PetscCall(PetscInfo(mat, "Check convertfrom (%s) -> %d\n", ((PetscObject)B)->type_name, !!conv));
4466: PetscCall(MatDestroy(&B));
4467: if (conv) goto foundconv;
4469: /* 4) See if a good general converter is known for the current matrix */
4470: if (mat->ops->convert) conv = mat->ops->convert;
4471: PetscCall(PetscInfo(mat, "Check general convert (%s) -> %d\n", ((PetscObject)mat)->type_name, !!conv));
4472: if (conv) goto foundconv;
4474: /* 5) Use a really basic converter. */
4475: PetscCall(PetscInfo(mat, "Using MatConvert_Basic\n"));
4476: conv = MatConvert_Basic;
4478: foundconv:
4479: PetscCall(PetscLogEventBegin(MAT_Convert, mat, 0, 0, 0));
4480: PetscCall((*conv)(mat, newtype, reuse, M));
4481: if (mat->rmap->mapping && mat->cmap->mapping && !(*M)->rmap->mapping && !(*M)->cmap->mapping) {
4482: /* the block sizes must be same if the mappings are copied over */
4483: (*M)->rmap->bs = mat->rmap->bs;
4484: (*M)->cmap->bs = mat->cmap->bs;
4485: PetscCall(PetscObjectReference((PetscObject)mat->rmap->mapping));
4486: PetscCall(PetscObjectReference((PetscObject)mat->cmap->mapping));
4487: (*M)->rmap->mapping = mat->rmap->mapping;
4488: (*M)->cmap->mapping = mat->cmap->mapping;
4489: }
4490: (*M)->stencil.dim = mat->stencil.dim;
4491: (*M)->stencil.noc = mat->stencil.noc;
4492: for (i = 0; i <= mat->stencil.dim + (mat->stencil.noc ? 0 : -1); i++) {
4493: (*M)->stencil.dims[i] = mat->stencil.dims[i];
4494: (*M)->stencil.starts[i] = mat->stencil.starts[i];
4495: }
4496: PetscCall(PetscLogEventEnd(MAT_Convert, mat, 0, 0, 0));
4497: }
4498: PetscCall(PetscObjectStateIncrease((PetscObject)*M));
4500: /* Copy Mat options */
4501: if (issymmetric == PETSC_BOOL3_TRUE) PetscCall(MatSetOption(*M, MAT_SYMMETRIC, PETSC_TRUE));
4502: else if (issymmetric == PETSC_BOOL3_FALSE) PetscCall(MatSetOption(*M, MAT_SYMMETRIC, PETSC_FALSE));
4503: if (ishermitian == PETSC_BOOL3_TRUE) PetscCall(MatSetOption(*M, MAT_HERMITIAN, PETSC_TRUE));
4504: else if (ishermitian == PETSC_BOOL3_FALSE) PetscCall(MatSetOption(*M, MAT_HERMITIAN, PETSC_FALSE));
4505: PetscFunctionReturn(PETSC_SUCCESS);
4506: }
4508: /*@
4509: MatFactorGetSolverType - Returns name of the package providing the factorization routines
4511: Not Collective
4513: Input Parameter:
4514: . mat - the matrix, must be a factored matrix
4516: Output Parameter:
4517: . type - the string name of the package (do not free this string)
4519: Level: intermediate
4521: Fortran Note:
4522: Pass in an empty string that is long enough and the package name will be copied into it.
4524: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSolverType`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`
4525: @*/
4526: PetscErrorCode MatFactorGetSolverType(Mat mat, MatSolverType *type)
4527: {
4528: PetscErrorCode (*conv)(Mat, MatSolverType *);
4530: PetscFunctionBegin;
4533: PetscAssertPointer(type, 2);
4534: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Only for factored matrix");
4535: PetscCall(PetscObjectQueryFunction((PetscObject)mat, "MatFactorGetSolverType_C", &conv));
4536: if (conv) PetscCall((*conv)(mat, type));
4537: else *type = MATSOLVERPETSC;
4538: PetscFunctionReturn(PETSC_SUCCESS);
4539: }
4541: typedef struct _MatSolverTypeForSpecifcType *MatSolverTypeForSpecifcType;
4542: struct _MatSolverTypeForSpecifcType {
4543: MatType mtype;
4544: /* no entry for MAT_FACTOR_NONE */
4545: PetscErrorCode (*createfactor[MAT_FACTOR_NUM_TYPES - 1])(Mat, MatFactorType, Mat *);
4546: MatSolverTypeForSpecifcType next;
4547: };
4549: typedef struct _MatSolverTypeHolder *MatSolverTypeHolder;
4550: struct _MatSolverTypeHolder {
4551: char *name;
4552: MatSolverTypeForSpecifcType handlers;
4553: MatSolverTypeHolder next;
4554: };
4556: static MatSolverTypeHolder MatSolverTypeHolders = NULL;
4558: /*@C
4559: MatSolverTypeRegister - Registers a `MatSolverType` that works for a particular matrix type
4561: Logically Collective, No Fortran Support
4563: Input Parameters:
4564: + package - name of the package, for example petsc or superlu
4565: . mtype - the matrix type that works with this package
4566: . ftype - the type of factorization supported by the package
4567: - createfactor - routine that will create the factored matrix ready to be used
4569: Level: developer
4571: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorGetSolverType()`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`,
4572: `MatGetFactor()`
4573: @*/
4574: PetscErrorCode MatSolverTypeRegister(MatSolverType package, MatType mtype, MatFactorType ftype, PetscErrorCode (*createfactor)(Mat, MatFactorType, Mat *))
4575: {
4576: MatSolverTypeHolder next = MatSolverTypeHolders, prev = NULL;
4577: PetscBool flg;
4578: MatSolverTypeForSpecifcType inext, iprev = NULL;
4580: PetscFunctionBegin;
4581: PetscCall(MatInitializePackage());
4582: if (!next) {
4583: PetscCall(PetscNew(&MatSolverTypeHolders));
4584: PetscCall(PetscStrallocpy(package, &MatSolverTypeHolders->name));
4585: PetscCall(PetscNew(&MatSolverTypeHolders->handlers));
4586: PetscCall(PetscStrallocpy(mtype, (char **)&MatSolverTypeHolders->handlers->mtype));
4587: MatSolverTypeHolders->handlers->createfactor[(int)ftype - 1] = createfactor;
4588: PetscFunctionReturn(PETSC_SUCCESS);
4589: }
4590: while (next) {
4591: PetscCall(PetscStrcasecmp(package, next->name, &flg));
4592: if (flg) {
4593: PetscCheck(next->handlers, PETSC_COMM_SELF, PETSC_ERR_PLIB, "MatSolverTypeHolder is missing handlers");
4594: inext = next->handlers;
4595: while (inext) {
4596: PetscCall(PetscStrcasecmp(mtype, inext->mtype, &flg));
4597: if (flg) {
4598: inext->createfactor[(int)ftype - 1] = createfactor;
4599: PetscFunctionReturn(PETSC_SUCCESS);
4600: }
4601: iprev = inext;
4602: inext = inext->next;
4603: }
4604: PetscCall(PetscNew(&iprev->next));
4605: PetscCall(PetscStrallocpy(mtype, (char **)&iprev->next->mtype));
4606: iprev->next->createfactor[(int)ftype - 1] = createfactor;
4607: PetscFunctionReturn(PETSC_SUCCESS);
4608: }
4609: prev = next;
4610: next = next->next;
4611: }
4612: PetscCall(PetscNew(&prev->next));
4613: PetscCall(PetscStrallocpy(package, &prev->next->name));
4614: PetscCall(PetscNew(&prev->next->handlers));
4615: PetscCall(PetscStrallocpy(mtype, (char **)&prev->next->handlers->mtype));
4616: prev->next->handlers->createfactor[(int)ftype - 1] = createfactor;
4617: PetscFunctionReturn(PETSC_SUCCESS);
4618: }
4620: /*@C
4621: MatSolverTypeGet - Gets the function that creates the factor matrix if it exist
4623: Input Parameters:
4624: + 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
4625: . ftype - the type of factorization supported by the type
4626: - mtype - the matrix type that works with this type
4628: Output Parameters:
4629: + foundtype - `PETSC_TRUE` if the type was registered
4630: . foundmtype - `PETSC_TRUE` if the type supports the requested mtype
4631: - createfactor - routine that will create the factored matrix ready to be used or `NULL` if not found
4633: Calling sequence of `createfactor`:
4634: + A - the matrix providing the factor matrix
4635: . ftype - the `MatFactorType` of the factor requested
4636: - B - the new factor matrix that responds to MatXXFactorSymbolic,Numeric() functions, such as `MatLUFactorSymbolic()`
4638: Level: developer
4640: Note:
4641: When `type` is `NULL` the available functions are searched for based on the order of the calls to `MatSolverTypeRegister()` in `MatInitializePackage()`.
4642: Since different PETSc configurations may have different external solvers, seemingly identical runs with different PETSc configurations may use a different solver.
4643: For example if one configuration had `--download-mumps` while a different one had `--download-superlu_dist`.
4645: .seealso: [](ch_matrices), `Mat`, `MatFactorType`, `MatType`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`, `MatSolverTypeRegister()`, `MatGetFactor()`,
4646: `MatInitializePackage()`
4647: @*/
4648: PetscErrorCode MatSolverTypeGet(MatSolverType type, MatType mtype, MatFactorType ftype, PetscBool *foundtype, PetscBool *foundmtype, PetscErrorCode (**createfactor)(Mat A, MatFactorType ftype, Mat *B))
4649: {
4650: MatSolverTypeHolder next = MatSolverTypeHolders;
4651: PetscBool flg;
4652: MatSolverTypeForSpecifcType inext;
4654: PetscFunctionBegin;
4655: if (foundtype) *foundtype = PETSC_FALSE;
4656: if (foundmtype) *foundmtype = PETSC_FALSE;
4657: if (createfactor) *createfactor = NULL;
4659: if (type) {
4660: while (next) {
4661: PetscCall(PetscStrcasecmp(type, next->name, &flg));
4662: if (flg) {
4663: if (foundtype) *foundtype = PETSC_TRUE;
4664: inext = next->handlers;
4665: while (inext) {
4666: PetscCall(PetscStrbeginswith(mtype, inext->mtype, &flg));
4667: if (flg) {
4668: if (foundmtype) *foundmtype = PETSC_TRUE;
4669: if (createfactor) *createfactor = inext->createfactor[(int)ftype - 1];
4670: PetscFunctionReturn(PETSC_SUCCESS);
4671: }
4672: inext = inext->next;
4673: }
4674: }
4675: next = next->next;
4676: }
4677: } else {
4678: while (next) {
4679: inext = next->handlers;
4680: while (inext) {
4681: PetscCall(PetscStrcmp(mtype, inext->mtype, &flg));
4682: if (flg && inext->createfactor[(int)ftype - 1]) {
4683: if (foundtype) *foundtype = PETSC_TRUE;
4684: if (foundmtype) *foundmtype = PETSC_TRUE;
4685: if (createfactor) *createfactor = inext->createfactor[(int)ftype - 1];
4686: PetscFunctionReturn(PETSC_SUCCESS);
4687: }
4688: inext = inext->next;
4689: }
4690: next = next->next;
4691: }
4692: /* try with base classes inext->mtype */
4693: next = MatSolverTypeHolders;
4694: while (next) {
4695: inext = next->handlers;
4696: while (inext) {
4697: PetscCall(PetscStrbeginswith(mtype, inext->mtype, &flg));
4698: if (flg && inext->createfactor[(int)ftype - 1]) {
4699: if (foundtype) *foundtype = PETSC_TRUE;
4700: if (foundmtype) *foundmtype = PETSC_TRUE;
4701: if (createfactor) *createfactor = inext->createfactor[(int)ftype - 1];
4702: PetscFunctionReturn(PETSC_SUCCESS);
4703: }
4704: inext = inext->next;
4705: }
4706: next = next->next;
4707: }
4708: }
4709: PetscFunctionReturn(PETSC_SUCCESS);
4710: }
4712: PetscErrorCode MatSolverTypeDestroy(void)
4713: {
4714: MatSolverTypeHolder next = MatSolverTypeHolders, prev;
4715: MatSolverTypeForSpecifcType inext, iprev;
4717: PetscFunctionBegin;
4718: while (next) {
4719: PetscCall(PetscFree(next->name));
4720: inext = next->handlers;
4721: while (inext) {
4722: PetscCall(PetscFree(inext->mtype));
4723: iprev = inext;
4724: inext = inext->next;
4725: PetscCall(PetscFree(iprev));
4726: }
4727: prev = next;
4728: next = next->next;
4729: PetscCall(PetscFree(prev));
4730: }
4731: MatSolverTypeHolders = NULL;
4732: PetscFunctionReturn(PETSC_SUCCESS);
4733: }
4735: /*@
4736: MatFactorGetCanUseOrdering - Indicates if the factorization can use the ordering provided in `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`
4738: Logically Collective
4740: Input Parameter:
4741: . mat - the matrix
4743: Output Parameter:
4744: . flg - `PETSC_TRUE` if uses the ordering
4746: Level: developer
4748: Note:
4749: Most internal PETSc factorizations use the ordering passed to the factorization routine but external
4750: packages do not, thus we want to skip generating the ordering when it is not needed or used.
4752: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`
4753: @*/
4754: PetscErrorCode MatFactorGetCanUseOrdering(Mat mat, PetscBool *flg)
4755: {
4756: PetscFunctionBegin;
4757: *flg = mat->canuseordering;
4758: PetscFunctionReturn(PETSC_SUCCESS);
4759: }
4761: /*@
4762: MatFactorGetPreferredOrdering - The preferred ordering for a particular matrix factor object
4764: Logically Collective
4766: Input Parameters:
4767: + mat - the matrix obtained with `MatGetFactor()`
4768: - ftype - the factorization type to be used
4770: Output Parameter:
4771: . otype - the preferred ordering type
4773: Level: developer
4775: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorType`, `MatOrderingType`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`
4776: @*/
4777: PetscErrorCode MatFactorGetPreferredOrdering(Mat mat, MatFactorType ftype, MatOrderingType *otype)
4778: {
4779: PetscFunctionBegin;
4780: *otype = mat->preferredordering[ftype];
4781: PetscCheck(*otype, PETSC_COMM_SELF, PETSC_ERR_PLIB, "MatFactor did not have a preferred ordering");
4782: PetscFunctionReturn(PETSC_SUCCESS);
4783: }
4785: /*@
4786: MatGetFactor - Returns a matrix suitable to calls to MatXXFactorSymbolic,Numeric()
4788: Collective
4790: Input Parameters:
4791: + mat - the matrix
4792: . type - name of solver type, for example, superlu, petsc (to use PETSc's solver if it is available), if this is 'NULL', then the first result that satisfies
4793: the other criteria is returned
4794: - ftype - factor type, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`
4796: Output Parameter:
4797: . f - the factor matrix used with MatXXFactorSymbolic,Numeric() calls. Can be `NULL` in some cases, see notes below.
4799: Options Database Keys:
4800: + -pc_factor_mat_solver_type <type> - choose the type at run time. When using `KSP` solvers
4801: - -mat_factor_bind_factorization <host, device> - Where to do matrix factorization? Default is device (might consume more device memory.
4802: One can choose host to save device memory). Currently only supported with `MATSEQAIJCUSPARSE` matrices.
4804: Level: intermediate
4806: Notes:
4807: The return matrix can be `NULL` if the requested factorization is not available, since some combinations of matrix types and factorization
4808: types registered with `MatSolverTypeRegister()` cannot be fully tested if not at runtime.
4810: Users usually access the factorization solvers via `KSP`
4812: Some PETSc matrix formats have alternative solvers available that are contained in alternative packages
4813: such as pastix, superlu, mumps etc. PETSc must have been ./configure to use the external solver, using the option --download-package or --with-package-dir
4815: When `type` is `NULL` the available results are searched for based on the order of the calls to `MatSolverTypeRegister()` in `MatInitializePackage()`.
4816: Since different PETSc configurations may have different external solvers, seemingly identical runs with different PETSc configurations may use a different solver.
4817: For example if one configuration had --download-mumps while a different one had --download-superlu_dist.
4819: Some of the packages have options for controlling the factorization, these are in the form -prefix_mat_packagename_packageoption
4820: where prefix is normally obtained from the calling `KSP`/`PC`. If `MatGetFactor()` is called directly one can set
4821: call `MatSetOptionsPrefixFactor()` on the originating matrix or `MatSetOptionsPrefix()` on the resulting factor matrix.
4823: Developer Note:
4824: This should actually be called `MatCreateFactor()` since it creates a new factor object
4826: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `KSP`, `MatSolverType`, `MatFactorType`, `MatCopy()`, `MatDuplicate()`,
4827: `MatGetFactorAvailable()`, `MatFactorGetCanUseOrdering()`, `MatSolverTypeRegister()`, `MatSolverTypeGet()`
4828: `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`, `MatInitializePackage()`
4829: @*/
4830: PetscErrorCode MatGetFactor(Mat mat, MatSolverType type, MatFactorType ftype, Mat *f)
4831: {
4832: PetscBool foundtype, foundmtype, shell, hasop = PETSC_FALSE;
4833: PetscErrorCode (*conv)(Mat, MatFactorType, Mat *);
4835: PetscFunctionBegin;
4839: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4840: MatCheckPreallocated(mat, 1);
4842: PetscCall(MatIsShell(mat, &shell));
4843: if (shell) PetscCall(MatHasOperation(mat, MATOP_GET_FACTOR, &hasop));
4844: if (hasop) {
4845: PetscUseTypeMethod(mat, getfactor, type, ftype, f);
4846: PetscFunctionReturn(PETSC_SUCCESS);
4847: }
4849: PetscCall(MatSolverTypeGet(type, ((PetscObject)mat)->type_name, ftype, &foundtype, &foundmtype, &conv));
4850: if (!foundtype) {
4851: if (type) {
4852: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_MISSING_FACTOR, "Could not locate solver type %s for factorization type %s and matrix type %s. Perhaps you must ./configure with --download-%s", type, MatFactorTypes[ftype],
4853: ((PetscObject)mat)->type_name, type);
4854: } else {
4855: SETERRQ(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);
4856: }
4857: }
4858: PetscCheck(foundmtype, PetscObjectComm((PetscObject)mat), PETSC_ERR_MISSING_FACTOR, "MatSolverType %s does not support matrix type %s", type, ((PetscObject)mat)->type_name);
4859: 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);
4861: PetscCall((*conv)(mat, ftype, f));
4862: if (mat->factorprefix) PetscCall(MatSetOptionsPrefix(*f, mat->factorprefix));
4863: PetscFunctionReturn(PETSC_SUCCESS);
4864: }
4866: /*@
4867: MatGetFactorAvailable - Returns a flag if matrix supports particular type and factor type
4869: Not Collective
4871: Input Parameters:
4872: + mat - the matrix
4873: . type - name of solver type, for example, superlu, petsc (to use PETSc's default)
4874: - ftype - factor type, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`
4876: Output Parameter:
4877: . flg - PETSC_TRUE if the factorization is available
4879: Level: intermediate
4881: Notes:
4882: Some PETSc matrix formats have alternative solvers available that are contained in alternative packages
4883: such as pastix, superlu, mumps etc.
4885: PETSc must have been ./configure to use the external solver, using the option --download-package
4887: Developer Note:
4888: This should actually be called `MatCreateFactorAvailable()` since `MatGetFactor()` creates a new factor object
4890: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatSolverType`, `MatFactorType`, `MatGetFactor()`, `MatCopy()`, `MatDuplicate()`, `MatSolverTypeRegister()`,
4891: `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`, `MatSolverTypeGet()`
4892: @*/
4893: PetscErrorCode MatGetFactorAvailable(Mat mat, MatSolverType type, MatFactorType ftype, PetscBool *flg)
4894: {
4895: PetscErrorCode (*gconv)(Mat, MatFactorType, Mat *);
4897: PetscFunctionBegin;
4899: PetscAssertPointer(flg, 4);
4901: *flg = PETSC_FALSE;
4902: if (!((PetscObject)mat)->type_name) PetscFunctionReturn(PETSC_SUCCESS);
4904: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4905: MatCheckPreallocated(mat, 1);
4907: PetscCall(MatSolverTypeGet(type, ((PetscObject)mat)->type_name, ftype, NULL, NULL, &gconv));
4908: *flg = gconv ? PETSC_TRUE : PETSC_FALSE;
4909: PetscFunctionReturn(PETSC_SUCCESS);
4910: }
4912: /*@
4913: MatDuplicate - Duplicates a matrix including the non-zero structure.
4915: Collective
4917: Input Parameters:
4918: + mat - the matrix
4919: - op - One of `MAT_DO_NOT_COPY_VALUES`, `MAT_COPY_VALUES`, or `MAT_SHARE_NONZERO_PATTERN`.
4920: See the manual page for `MatDuplicateOption()` for an explanation of these options.
4922: Output Parameter:
4923: . M - pointer to place new matrix
4925: Level: intermediate
4927: Notes:
4928: You cannot change the nonzero pattern for the parent or child matrix later if you use `MAT_SHARE_NONZERO_PATTERN`.
4930: If `op` is not `MAT_COPY_VALUES` the numerical values in the new matrix are zeroed.
4932: 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.
4934: When original mat is a product of matrix operation, e.g., an output of `MatMatMult()` or `MatCreateSubMatrix()`, only the matrix data structure of `mat`
4935: is duplicated and the internal data structures created for the reuse of previous matrix operations are not duplicated.
4936: User should not use `MatDuplicate()` to create new matrix `M` if `M` is intended to be reused as the product of matrix operation.
4938: .seealso: [](ch_matrices), `Mat`, `MatCopy()`, `MatConvert()`, `MatDuplicateOption`
4939: @*/
4940: PetscErrorCode MatDuplicate(Mat mat, MatDuplicateOption op, Mat *M)
4941: {
4942: Mat B;
4943: VecType vtype;
4944: PetscInt i;
4945: PetscObject dm, container_h, container_d;
4946: void (*viewf)(void);
4948: PetscFunctionBegin;
4951: PetscAssertPointer(M, 3);
4952: PetscCheck(op != MAT_COPY_VALUES || mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "MAT_COPY_VALUES not allowed for unassembled matrix");
4953: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4954: MatCheckPreallocated(mat, 1);
4956: *M = NULL;
4957: PetscCall(PetscLogEventBegin(MAT_Convert, mat, 0, 0, 0));
4958: PetscUseTypeMethod(mat, duplicate, op, M);
4959: PetscCall(PetscLogEventEnd(MAT_Convert, mat, 0, 0, 0));
4960: B = *M;
4962: PetscCall(MatGetOperation(mat, MATOP_VIEW, &viewf));
4963: if (viewf) PetscCall(MatSetOperation(B, MATOP_VIEW, viewf));
4964: PetscCall(MatGetVecType(mat, &vtype));
4965: PetscCall(MatSetVecType(B, vtype));
4967: B->stencil.dim = mat->stencil.dim;
4968: B->stencil.noc = mat->stencil.noc;
4969: for (i = 0; i <= mat->stencil.dim + (mat->stencil.noc ? 0 : -1); i++) {
4970: B->stencil.dims[i] = mat->stencil.dims[i];
4971: B->stencil.starts[i] = mat->stencil.starts[i];
4972: }
4974: B->nooffproczerorows = mat->nooffproczerorows;
4975: B->nooffprocentries = mat->nooffprocentries;
4977: PetscCall(PetscObjectQuery((PetscObject)mat, "__PETSc_dm", &dm));
4978: if (dm) PetscCall(PetscObjectCompose((PetscObject)B, "__PETSc_dm", dm));
4979: PetscCall(PetscObjectQuery((PetscObject)mat, "__PETSc_MatCOOStruct_Host", &container_h));
4980: if (container_h) PetscCall(PetscObjectCompose((PetscObject)B, "__PETSc_MatCOOStruct_Host", container_h));
4981: PetscCall(PetscObjectQuery((PetscObject)mat, "__PETSc_MatCOOStruct_Device", &container_d));
4982: if (container_d) PetscCall(PetscObjectCompose((PetscObject)B, "__PETSc_MatCOOStruct_Device", container_d));
4983: if (op == MAT_COPY_VALUES) PetscCall(MatPropagateSymmetryOptions(mat, B));
4984: PetscCall(PetscObjectStateIncrease((PetscObject)B));
4985: PetscFunctionReturn(PETSC_SUCCESS);
4986: }
4988: /*@
4989: MatGetDiagonal - Gets the diagonal of a matrix as a `Vec`
4991: Logically Collective
4993: Input Parameter:
4994: . mat - the matrix
4996: Output Parameter:
4997: . v - the diagonal of the matrix
4999: Level: intermediate
5001: Note:
5002: If `mat` has local sizes `n` x `m`, this routine fills the first `ndiag = min(n, m)` entries
5003: of `v` with the diagonal values. Thus `v` must have local size of at least `ndiag`. If `v`
5004: is larger than `ndiag`, the values of the remaining entries are unspecified.
5006: Currently only correct in parallel for square matrices.
5008: .seealso: [](ch_matrices), `Mat`, `Vec`, `MatGetRow()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMaxAbs()`
5009: @*/
5010: PetscErrorCode MatGetDiagonal(Mat mat, Vec v)
5011: {
5012: PetscFunctionBegin;
5016: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5017: MatCheckPreallocated(mat, 1);
5018: if (PetscDefined(USE_DEBUG)) {
5019: PetscInt nv, row, col, ndiag;
5021: PetscCall(VecGetLocalSize(v, &nv));
5022: PetscCall(MatGetLocalSize(mat, &row, &col));
5023: ndiag = PetscMin(row, col);
5024: 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);
5025: }
5027: PetscUseTypeMethod(mat, getdiagonal, v);
5028: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5029: PetscFunctionReturn(PETSC_SUCCESS);
5030: }
5032: /*@
5033: MatGetRowMin - Gets the minimum value (of the real part) of each
5034: row of the matrix
5036: Logically Collective
5038: Input Parameter:
5039: . mat - the matrix
5041: Output Parameters:
5042: + v - the vector for storing the maximums
5043: - idx - the indices of the column found for each row (optional, pass `NULL` if not needed)
5045: Level: intermediate
5047: Note:
5048: The result of this call are the same as if one converted the matrix to dense format
5049: and found the minimum value in each row (i.e. the implicit zeros are counted as zeros).
5051: This code is only implemented for a couple of matrix formats.
5053: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMaxAbs()`, `MatGetRowMinAbs()`,
5054: `MatGetRowMax()`
5055: @*/
5056: PetscErrorCode MatGetRowMin(Mat mat, Vec v, PetscInt idx[])
5057: {
5058: PetscFunctionBegin;
5062: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5064: if (!mat->cmap->N) {
5065: PetscCall(VecSet(v, PETSC_MAX_REAL));
5066: if (idx) {
5067: PetscInt i, m = mat->rmap->n;
5068: for (i = 0; i < m; i++) idx[i] = -1;
5069: }
5070: } else {
5071: MatCheckPreallocated(mat, 1);
5072: }
5073: PetscUseTypeMethod(mat, getrowmin, v, idx);
5074: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5075: PetscFunctionReturn(PETSC_SUCCESS);
5076: }
5078: /*@
5079: MatGetRowMinAbs - Gets the minimum value (in absolute value) of each
5080: row of the matrix
5082: Logically Collective
5084: Input Parameter:
5085: . mat - the matrix
5087: Output Parameters:
5088: + v - the vector for storing the minimums
5089: - idx - the indices of the column found for each row (or `NULL` if not needed)
5091: Level: intermediate
5093: Notes:
5094: if a row is completely empty or has only 0.0 values, then the `idx` value for that
5095: row is 0 (the first column).
5097: This code is only implemented for a couple of matrix formats.
5099: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMax()`, `MatGetRowMaxAbs()`, `MatGetRowMin()`
5100: @*/
5101: PetscErrorCode MatGetRowMinAbs(Mat mat, Vec v, PetscInt idx[])
5102: {
5103: PetscFunctionBegin;
5107: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5108: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5110: if (!mat->cmap->N) {
5111: PetscCall(VecSet(v, 0.0));
5112: if (idx) {
5113: PetscInt i, m = mat->rmap->n;
5114: for (i = 0; i < m; i++) idx[i] = -1;
5115: }
5116: } else {
5117: MatCheckPreallocated(mat, 1);
5118: if (idx) PetscCall(PetscArrayzero(idx, mat->rmap->n));
5119: PetscUseTypeMethod(mat, getrowminabs, v, idx);
5120: }
5121: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5122: PetscFunctionReturn(PETSC_SUCCESS);
5123: }
5125: /*@
5126: MatGetRowMax - Gets the maximum value (of the real part) of each
5127: row of the matrix
5129: Logically Collective
5131: Input Parameter:
5132: . mat - the matrix
5134: Output Parameters:
5135: + v - the vector for storing the maximums
5136: - idx - the indices of the column found for each row (optional, otherwise pass `NULL`)
5138: Level: intermediate
5140: Notes:
5141: The result of this call are the same as if one converted the matrix to dense format
5142: and found the minimum value in each row (i.e. the implicit zeros are counted as zeros).
5144: This code is only implemented for a couple of matrix formats.
5146: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMaxAbs()`, `MatGetRowMin()`, `MatGetRowMinAbs()`
5147: @*/
5148: PetscErrorCode MatGetRowMax(Mat mat, Vec v, PetscInt idx[])
5149: {
5150: PetscFunctionBegin;
5154: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5156: if (!mat->cmap->N) {
5157: PetscCall(VecSet(v, PETSC_MIN_REAL));
5158: if (idx) {
5159: PetscInt i, m = mat->rmap->n;
5160: for (i = 0; i < m; i++) idx[i] = -1;
5161: }
5162: } else {
5163: MatCheckPreallocated(mat, 1);
5164: PetscUseTypeMethod(mat, getrowmax, v, idx);
5165: }
5166: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5167: PetscFunctionReturn(PETSC_SUCCESS);
5168: }
5170: /*@
5171: MatGetRowMaxAbs - Gets the maximum value (in absolute value) of each
5172: row of the matrix
5174: Logically Collective
5176: Input Parameter:
5177: . mat - the matrix
5179: Output Parameters:
5180: + v - the vector for storing the maximums
5181: - idx - the indices of the column found for each row (or `NULL` if not needed)
5183: Level: intermediate
5185: Notes:
5186: if a row is completely empty or has only 0.0 values, then the `idx` value for that
5187: row is 0 (the first column).
5189: This code is only implemented for a couple of matrix formats.
5191: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowSum()`, `MatGetRowMin()`, `MatGetRowMinAbs()`
5192: @*/
5193: PetscErrorCode MatGetRowMaxAbs(Mat mat, Vec v, PetscInt idx[])
5194: {
5195: PetscFunctionBegin;
5199: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5201: if (!mat->cmap->N) {
5202: PetscCall(VecSet(v, 0.0));
5203: if (idx) {
5204: PetscInt i, m = mat->rmap->n;
5205: for (i = 0; i < m; i++) idx[i] = -1;
5206: }
5207: } else {
5208: MatCheckPreallocated(mat, 1);
5209: if (idx) PetscCall(PetscArrayzero(idx, mat->rmap->n));
5210: PetscUseTypeMethod(mat, getrowmaxabs, v, idx);
5211: }
5212: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5213: PetscFunctionReturn(PETSC_SUCCESS);
5214: }
5216: /*@
5217: MatGetRowSumAbs - Gets the sum value (in absolute value) of each row of the matrix
5219: Logically Collective
5221: Input Parameter:
5222: . mat - the matrix
5224: Output Parameter:
5225: . v - the vector for storing the sum
5227: Level: intermediate
5229: This code is only implemented for a couple of matrix formats.
5231: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMax()`, `MatGetRowMin()`, `MatGetRowMinAbs()`
5232: @*/
5233: PetscErrorCode MatGetRowSumAbs(Mat mat, Vec v)
5234: {
5235: PetscFunctionBegin;
5239: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5241: if (!mat->cmap->N) {
5242: PetscCall(VecSet(v, 0.0));
5243: } else {
5244: MatCheckPreallocated(mat, 1);
5245: PetscUseTypeMethod(mat, getrowsumabs, v);
5246: }
5247: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5248: PetscFunctionReturn(PETSC_SUCCESS);
5249: }
5251: /*@
5252: MatGetRowSum - Gets the sum of each row of the matrix
5254: Logically or Neighborhood Collective
5256: Input Parameter:
5257: . mat - the matrix
5259: Output Parameter:
5260: . v - the vector for storing the sum of rows
5262: Level: intermediate
5264: Note:
5265: This code is slow since it is not currently specialized for different formats
5267: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMax()`, `MatGetRowMin()`, `MatGetRowMaxAbs()`, `MatGetRowMinAbs()`, `MatGetRowSumAbs()`
5268: @*/
5269: PetscErrorCode MatGetRowSum(Mat mat, Vec v)
5270: {
5271: Vec ones;
5273: PetscFunctionBegin;
5277: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5278: MatCheckPreallocated(mat, 1);
5279: PetscCall(MatCreateVecs(mat, &ones, NULL));
5280: PetscCall(VecSet(ones, 1.));
5281: PetscCall(MatMult(mat, ones, v));
5282: PetscCall(VecDestroy(&ones));
5283: PetscFunctionReturn(PETSC_SUCCESS);
5284: }
5286: /*@
5287: MatTransposeSetPrecursor - Set the matrix from which the second matrix will receive numerical transpose data with a call to `MatTranspose`(A,`MAT_REUSE_MATRIX`,&B)
5288: when B was not obtained with `MatTranspose`(A,`MAT_INITIAL_MATRIX`,&B)
5290: Collective
5292: Input Parameter:
5293: . mat - the matrix to provide the transpose
5295: Output Parameter:
5296: . 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
5298: Level: advanced
5300: Note:
5301: 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
5302: routine allows bypassing that call.
5304: .seealso: [](ch_matrices), `Mat`, `MatTransposeSymbolic()`, `MatTranspose()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`
5305: @*/
5306: PetscErrorCode MatTransposeSetPrecursor(Mat mat, Mat B)
5307: {
5308: MatParentState *rb = NULL;
5310: PetscFunctionBegin;
5311: PetscCall(PetscNew(&rb));
5312: rb->id = ((PetscObject)mat)->id;
5313: rb->state = 0;
5314: PetscCall(MatGetNonzeroState(mat, &rb->nonzerostate));
5315: PetscCall(PetscObjectContainerCompose((PetscObject)B, "MatTransposeParent", rb, PetscContainerUserDestroyDefault));
5316: PetscFunctionReturn(PETSC_SUCCESS);
5317: }
5319: /*@
5320: MatTranspose - Computes the transpose of a matrix, either in-place or out-of-place.
5322: Collective
5324: Input Parameters:
5325: + mat - the matrix to transpose
5326: - reuse - either `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, or `MAT_INPLACE_MATRIX`
5328: Output Parameter:
5329: . B - the transpose of the matrix
5331: Level: intermediate
5333: Notes:
5334: If you use `MAT_INPLACE_MATRIX` then you must pass in `&mat` for `B`
5336: `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
5337: transpose, call `MatTransposeSetPrecursor(mat, B)` before calling this routine.
5339: 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.
5341: Consider using `MatCreateTranspose()` instead if you only need a matrix that behaves like the transpose but don't need the storage to be changed.
5342: For example, the result of `MatCreateTranspose()` will compute the transpose of the given matrix times a vector for matrix-vector products computed with `MatMult()`.
5344: If `mat` is unchanged from the last call this function returns immediately without recomputing the result
5346: If you only need the symbolic transpose of a matrix, and not the numerical values, use `MatTransposeSymbolic()`
5348: .seealso: [](ch_matrices), `Mat`, `MatTransposeSetPrecursor()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`,
5349: `MatTransposeSymbolic()`, `MatCreateTranspose()`
5350: @*/
5351: PetscErrorCode MatTranspose(Mat mat, MatReuse reuse, Mat *B)
5352: {
5353: PetscContainer rB = NULL;
5354: MatParentState *rb = NULL;
5356: PetscFunctionBegin;
5359: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5360: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5361: PetscCheck(reuse != MAT_INPLACE_MATRIX || mat == *B, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "MAT_INPLACE_MATRIX requires last matrix to match first");
5362: PetscCheck(reuse != MAT_REUSE_MATRIX || mat != *B, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Perhaps you mean MAT_INPLACE_MATRIX");
5363: MatCheckPreallocated(mat, 1);
5364: if (reuse == MAT_REUSE_MATRIX) {
5365: PetscCall(PetscObjectQuery((PetscObject)*B, "MatTransposeParent", (PetscObject *)&rB));
5366: PetscCheck(rB, PetscObjectComm((PetscObject)*B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from call to MatTranspose(). Suggest MatTransposeSetPrecursor().");
5367: PetscCall(PetscContainerGetPointer(rB, (void **)&rb));
5368: PetscCheck(rb->id == ((PetscObject)mat)->id, PetscObjectComm((PetscObject)*B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from input matrix");
5369: if (rb->state == ((PetscObject)mat)->state) PetscFunctionReturn(PETSC_SUCCESS);
5370: }
5372: PetscCall(PetscLogEventBegin(MAT_Transpose, mat, 0, 0, 0));
5373: if (reuse != MAT_INPLACE_MATRIX || mat->symmetric != PETSC_BOOL3_TRUE) {
5374: PetscUseTypeMethod(mat, transpose, reuse, B);
5375: PetscCall(PetscObjectStateIncrease((PetscObject)*B));
5376: }
5377: PetscCall(PetscLogEventEnd(MAT_Transpose, mat, 0, 0, 0));
5379: if (reuse == MAT_INITIAL_MATRIX) PetscCall(MatTransposeSetPrecursor(mat, *B));
5380: if (reuse != MAT_INPLACE_MATRIX) {
5381: PetscCall(PetscObjectQuery((PetscObject)*B, "MatTransposeParent", (PetscObject *)&rB));
5382: PetscCall(PetscContainerGetPointer(rB, (void **)&rb));
5383: rb->state = ((PetscObject)mat)->state;
5384: rb->nonzerostate = mat->nonzerostate;
5385: }
5386: PetscFunctionReturn(PETSC_SUCCESS);
5387: }
5389: /*@
5390: MatTransposeSymbolic - Computes the symbolic part of the transpose of a matrix.
5392: Collective
5394: Input Parameter:
5395: . A - the matrix to transpose
5397: Output Parameter:
5398: . 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
5399: numerical portion.
5401: Level: intermediate
5403: Note:
5404: This is not supported for many matrix types, use `MatTranspose()` in those cases
5406: .seealso: [](ch_matrices), `Mat`, `MatTransposeSetPrecursor()`, `MatTranspose()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`
5407: @*/
5408: PetscErrorCode MatTransposeSymbolic(Mat A, Mat *B)
5409: {
5410: PetscFunctionBegin;
5413: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5414: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5415: PetscCall(PetscLogEventBegin(MAT_Transpose, A, 0, 0, 0));
5416: PetscUseTypeMethod(A, transposesymbolic, B);
5417: PetscCall(PetscLogEventEnd(MAT_Transpose, A, 0, 0, 0));
5419: PetscCall(MatTransposeSetPrecursor(A, *B));
5420: PetscFunctionReturn(PETSC_SUCCESS);
5421: }
5423: PetscErrorCode MatTransposeCheckNonzeroState_Private(Mat A, Mat B)
5424: {
5425: PetscContainer rB;
5426: MatParentState *rb;
5428: PetscFunctionBegin;
5431: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5432: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5433: PetscCall(PetscObjectQuery((PetscObject)B, "MatTransposeParent", (PetscObject *)&rB));
5434: PetscCheck(rB, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from call to MatTranspose()");
5435: PetscCall(PetscContainerGetPointer(rB, (void **)&rb));
5436: PetscCheck(rb->id == ((PetscObject)A)->id, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from input matrix");
5437: PetscCheck(rb->nonzerostate == A->nonzerostate, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONGSTATE, "Reuse matrix has changed nonzero structure");
5438: PetscFunctionReturn(PETSC_SUCCESS);
5439: }
5441: /*@
5442: MatIsTranspose - Test whether a matrix is another one's transpose,
5443: or its own, in which case it tests symmetry.
5445: Collective
5447: Input Parameters:
5448: + A - the matrix to test
5449: . B - the matrix to test against, this can equal the first parameter
5450: - tol - tolerance, differences between entries smaller than this are counted as zero
5452: Output Parameter:
5453: . flg - the result
5455: Level: intermediate
5457: Notes:
5458: The sequential algorithm has a running time of the order of the number of nonzeros; the parallel
5459: test involves parallel copies of the block off-diagonal parts of the matrix.
5461: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsSymmetric()`, `MatIsHermitian()`
5462: @*/
5463: PetscErrorCode MatIsTranspose(Mat A, Mat B, PetscReal tol, PetscBool *flg)
5464: {
5465: PetscErrorCode (*f)(Mat, Mat, PetscReal, PetscBool *), (*g)(Mat, Mat, PetscReal, PetscBool *);
5467: PetscFunctionBegin;
5470: PetscAssertPointer(flg, 4);
5471: PetscCall(PetscObjectQueryFunction((PetscObject)A, "MatIsTranspose_C", &f));
5472: PetscCall(PetscObjectQueryFunction((PetscObject)B, "MatIsTranspose_C", &g));
5473: *flg = PETSC_FALSE;
5474: if (f && g) {
5475: PetscCheck(f == g, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_NOTSAMETYPE, "Matrices do not have the same comparator for symmetry test");
5476: PetscCall((*f)(A, B, tol, flg));
5477: } else {
5478: MatType mattype;
5480: PetscCall(MatGetType(f ? B : A, &mattype));
5481: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Matrix of type %s does not support checking for transpose", mattype);
5482: }
5483: PetscFunctionReturn(PETSC_SUCCESS);
5484: }
5486: /*@
5487: MatHermitianTranspose - Computes an in-place or out-of-place Hermitian transpose of a matrix in complex conjugate.
5489: Collective
5491: Input Parameters:
5492: + mat - the matrix to transpose and complex conjugate
5493: - reuse - either `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, or `MAT_INPLACE_MATRIX`
5495: Output Parameter:
5496: . B - the Hermitian transpose
5498: Level: intermediate
5500: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`
5501: @*/
5502: PetscErrorCode MatHermitianTranspose(Mat mat, MatReuse reuse, Mat *B)
5503: {
5504: PetscFunctionBegin;
5505: PetscCall(MatTranspose(mat, reuse, B));
5506: #if defined(PETSC_USE_COMPLEX)
5507: PetscCall(MatConjugate(*B));
5508: #endif
5509: PetscFunctionReturn(PETSC_SUCCESS);
5510: }
5512: /*@
5513: MatIsHermitianTranspose - Test whether a matrix is another one's Hermitian transpose,
5515: Collective
5517: Input Parameters:
5518: + A - the matrix to test
5519: . B - the matrix to test against, this can equal the first parameter
5520: - tol - tolerance, differences between entries smaller than this are counted as zero
5522: Output Parameter:
5523: . flg - the result
5525: Level: intermediate
5527: Notes:
5528: Only available for `MATAIJ` matrices.
5530: The sequential algorithm
5531: has a running time of the order of the number of nonzeros; the parallel
5532: test involves parallel copies of the block off-diagonal parts of the matrix.
5534: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsSymmetric()`, `MatIsHermitian()`, `MatIsTranspose()`
5535: @*/
5536: PetscErrorCode MatIsHermitianTranspose(Mat A, Mat B, PetscReal tol, PetscBool *flg)
5537: {
5538: PetscErrorCode (*f)(Mat, Mat, PetscReal, PetscBool *), (*g)(Mat, Mat, PetscReal, PetscBool *);
5540: PetscFunctionBegin;
5543: PetscAssertPointer(flg, 4);
5544: PetscCall(PetscObjectQueryFunction((PetscObject)A, "MatIsHermitianTranspose_C", &f));
5545: PetscCall(PetscObjectQueryFunction((PetscObject)B, "MatIsHermitianTranspose_C", &g));
5546: if (f && g) {
5547: PetscCheck(f != g, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_NOTSAMETYPE, "Matrices do not have the same comparator for Hermitian test");
5548: PetscCall((*f)(A, B, tol, flg));
5549: }
5550: PetscFunctionReturn(PETSC_SUCCESS);
5551: }
5553: /*@
5554: MatPermute - Creates a new matrix with rows and columns permuted from the
5555: original.
5557: Collective
5559: Input Parameters:
5560: + mat - the matrix to permute
5561: . row - row permutation, each processor supplies only the permutation for its rows
5562: - col - column permutation, each processor supplies only the permutation for its columns
5564: Output Parameter:
5565: . B - the permuted matrix
5567: Level: advanced
5569: Note:
5570: The index sets map from row/col of permuted matrix to row/col of original matrix.
5571: The index sets should be on the same communicator as mat and have the same local sizes.
5573: Developer Note:
5574: If you want to implement `MatPermute()` for a matrix type, and your approach doesn't
5575: exploit the fact that row and col are permutations, consider implementing the
5576: more general `MatCreateSubMatrix()` instead.
5578: .seealso: [](ch_matrices), `Mat`, `MatGetOrdering()`, `ISAllGather()`, `MatCreateSubMatrix()`
5579: @*/
5580: PetscErrorCode MatPermute(Mat mat, IS row, IS col, Mat *B)
5581: {
5582: PetscFunctionBegin;
5587: PetscAssertPointer(B, 4);
5588: PetscCheckSameComm(mat, 1, row, 2);
5589: if (row != col) PetscCheckSameComm(row, 2, col, 3);
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(mat->ops->permute || mat->ops->createsubmatrix, PETSC_COMM_SELF, PETSC_ERR_SUP, "MatPermute not available for Mat type %s", ((PetscObject)mat)->type_name);
5593: MatCheckPreallocated(mat, 1);
5595: if (mat->ops->permute) {
5596: PetscUseTypeMethod(mat, permute, row, col, B);
5597: PetscCall(PetscObjectStateIncrease((PetscObject)*B));
5598: } else {
5599: PetscCall(MatCreateSubMatrix(mat, row, col, MAT_INITIAL_MATRIX, B));
5600: }
5601: PetscFunctionReturn(PETSC_SUCCESS);
5602: }
5604: /*@
5605: MatEqual - Compares two matrices.
5607: Collective
5609: Input Parameters:
5610: + A - the first matrix
5611: - B - the second matrix
5613: Output Parameter:
5614: . flg - `PETSC_TRUE` if the matrices are equal; `PETSC_FALSE` otherwise.
5616: Level: intermediate
5618: .seealso: [](ch_matrices), `Mat`
5619: @*/
5620: PetscErrorCode MatEqual(Mat A, Mat B, PetscBool *flg)
5621: {
5622: PetscFunctionBegin;
5627: PetscAssertPointer(flg, 3);
5628: PetscCheckSameComm(A, 1, B, 2);
5629: MatCheckPreallocated(A, 1);
5630: MatCheckPreallocated(B, 2);
5631: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5632: PetscCheck(B->assembled, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5633: 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,
5634: B->cmap->N);
5635: if (A->ops->equal && A->ops->equal == B->ops->equal) {
5636: PetscUseTypeMethod(A, equal, B, flg);
5637: } else {
5638: PetscCall(MatMultEqual(A, B, 10, flg));
5639: }
5640: PetscFunctionReturn(PETSC_SUCCESS);
5641: }
5643: /*@
5644: MatDiagonalScale - Scales a matrix on the left and right by diagonal
5645: matrices that are stored as vectors. Either of the two scaling
5646: matrices can be `NULL`.
5648: Collective
5650: Input Parameters:
5651: + mat - the matrix to be scaled
5652: . l - the left scaling vector (or `NULL`)
5653: - r - the right scaling vector (or `NULL`)
5655: Level: intermediate
5657: Note:
5658: `MatDiagonalScale()` computes $A = LAR$, where
5659: L = a diagonal matrix (stored as a vector), R = a diagonal matrix (stored as a vector)
5660: The L scales the rows of the matrix, the R scales the columns of the matrix.
5662: .seealso: [](ch_matrices), `Mat`, `MatScale()`, `MatShift()`, `MatDiagonalSet()`
5663: @*/
5664: PetscErrorCode MatDiagonalScale(Mat mat, Vec l, Vec r)
5665: {
5666: PetscFunctionBegin;
5669: if (l) {
5671: PetscCheckSameComm(mat, 1, l, 2);
5672: }
5673: if (r) {
5675: PetscCheckSameComm(mat, 1, r, 3);
5676: }
5677: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5678: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5679: MatCheckPreallocated(mat, 1);
5680: if (!l && !r) PetscFunctionReturn(PETSC_SUCCESS);
5682: PetscCall(PetscLogEventBegin(MAT_Scale, mat, 0, 0, 0));
5683: PetscUseTypeMethod(mat, diagonalscale, l, r);
5684: PetscCall(PetscLogEventEnd(MAT_Scale, mat, 0, 0, 0));
5685: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
5686: if (l != r) mat->symmetric = PETSC_BOOL3_FALSE;
5687: PetscFunctionReturn(PETSC_SUCCESS);
5688: }
5690: /*@
5691: MatScale - Scales all elements of a matrix by a given number.
5693: Logically Collective
5695: Input Parameters:
5696: + mat - the matrix to be scaled
5697: - a - the scaling value
5699: Level: intermediate
5701: .seealso: [](ch_matrices), `Mat`, `MatDiagonalScale()`
5702: @*/
5703: PetscErrorCode MatScale(Mat mat, PetscScalar a)
5704: {
5705: PetscFunctionBegin;
5708: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5709: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5711: MatCheckPreallocated(mat, 1);
5713: PetscCall(PetscLogEventBegin(MAT_Scale, mat, 0, 0, 0));
5714: if (a != (PetscScalar)1.0) {
5715: PetscUseTypeMethod(mat, scale, a);
5716: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
5717: }
5718: PetscCall(PetscLogEventEnd(MAT_Scale, mat, 0, 0, 0));
5719: PetscFunctionReturn(PETSC_SUCCESS);
5720: }
5722: /*@
5723: MatNorm - Calculates various norms of a matrix.
5725: Collective
5727: Input Parameters:
5728: + mat - the matrix
5729: - type - the type of norm, `NORM_1`, `NORM_FROBENIUS`, `NORM_INFINITY`
5731: Output Parameter:
5732: . nrm - the resulting norm
5734: Level: intermediate
5736: .seealso: [](ch_matrices), `Mat`
5737: @*/
5738: PetscErrorCode MatNorm(Mat mat, NormType type, PetscReal *nrm)
5739: {
5740: PetscFunctionBegin;
5743: PetscAssertPointer(nrm, 3);
5745: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5746: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5747: MatCheckPreallocated(mat, 1);
5749: PetscUseTypeMethod(mat, norm, type, nrm);
5750: PetscFunctionReturn(PETSC_SUCCESS);
5751: }
5753: /*
5754: This variable is used to prevent counting of MatAssemblyBegin() that
5755: are called from within a MatAssemblyEnd().
5756: */
5757: static PetscInt MatAssemblyEnd_InUse = 0;
5758: /*@
5759: MatAssemblyBegin - Begins assembling the matrix. This routine should
5760: be called after completing all calls to `MatSetValues()`.
5762: Collective
5764: Input Parameters:
5765: + mat - the matrix
5766: - type - type of assembly, either `MAT_FLUSH_ASSEMBLY` or `MAT_FINAL_ASSEMBLY`
5768: Level: beginner
5770: Notes:
5771: `MatSetValues()` generally caches the values that belong to other MPI processes. The matrix is ready to
5772: use only after `MatAssemblyBegin()` and `MatAssemblyEnd()` have been called.
5774: Use `MAT_FLUSH_ASSEMBLY` when switching between `ADD_VALUES` and `INSERT_VALUES`
5775: in `MatSetValues()`; use `MAT_FINAL_ASSEMBLY` for the final assembly before
5776: using the matrix.
5778: ALL processes that share a matrix MUST call `MatAssemblyBegin()` and `MatAssemblyEnd()` the SAME NUMBER of times, and each time with the
5779: 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
5780: a global collective operation requiring all processes that share the matrix.
5782: Space for preallocated nonzeros that is not filled by a call to `MatSetValues()` or a related routine are compressed
5783: out by assembly. If you intend to use that extra space on a subsequent assembly, be sure to insert explicit zeros
5784: before `MAT_FINAL_ASSEMBLY` so the space is not compressed out.
5786: .seealso: [](ch_matrices), `Mat`, `MatAssemblyEnd()`, `MatSetValues()`, `MatAssembled()`
5787: @*/
5788: PetscErrorCode MatAssemblyBegin(Mat mat, MatAssemblyType type)
5789: {
5790: PetscFunctionBegin;
5793: MatCheckPreallocated(mat, 1);
5794: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix. Did you forget to call MatSetUnfactored()?");
5795: if (mat->assembled) {
5796: mat->was_assembled = PETSC_TRUE;
5797: mat->assembled = PETSC_FALSE;
5798: }
5800: if (!MatAssemblyEnd_InUse) {
5801: PetscCall(PetscLogEventBegin(MAT_AssemblyBegin, mat, 0, 0, 0));
5802: PetscTryTypeMethod(mat, assemblybegin, type);
5803: PetscCall(PetscLogEventEnd(MAT_AssemblyBegin, mat, 0, 0, 0));
5804: } else PetscTryTypeMethod(mat, assemblybegin, type);
5805: PetscFunctionReturn(PETSC_SUCCESS);
5806: }
5808: /*@
5809: MatAssembled - Indicates if a matrix has been assembled and is ready for
5810: use; for example, in matrix-vector product.
5812: Not Collective
5814: Input Parameter:
5815: . mat - the matrix
5817: Output Parameter:
5818: . assembled - `PETSC_TRUE` or `PETSC_FALSE`
5820: Level: advanced
5822: .seealso: [](ch_matrices), `Mat`, `MatAssemblyEnd()`, `MatSetValues()`, `MatAssemblyBegin()`
5823: @*/
5824: PetscErrorCode MatAssembled(Mat mat, PetscBool *assembled)
5825: {
5826: PetscFunctionBegin;
5828: PetscAssertPointer(assembled, 2);
5829: *assembled = mat->assembled;
5830: PetscFunctionReturn(PETSC_SUCCESS);
5831: }
5833: /*@
5834: MatAssemblyEnd - Completes assembling the matrix. This routine should
5835: be called after `MatAssemblyBegin()`.
5837: Collective
5839: Input Parameters:
5840: + mat - the matrix
5841: - type - type of assembly, either `MAT_FLUSH_ASSEMBLY` or `MAT_FINAL_ASSEMBLY`
5843: Options Database Keys:
5844: + -mat_view ::ascii_info - Prints info on matrix at conclusion of `MatAssemblyEnd()`
5845: . -mat_view ::ascii_info_detail - Prints more detailed info
5846: . -mat_view - Prints matrix in ASCII format
5847: . -mat_view ::ascii_matlab - Prints matrix in MATLAB format
5848: . -mat_view draw - draws nonzero structure of matrix, using `MatView()` and `PetscDrawOpenX()`.
5849: . -display <name> - Sets display name (default is host)
5850: . -draw_pause <sec> - Sets number of seconds to pause after display
5851: . -mat_view socket - Sends matrix to socket, can be accessed from MATLAB (See [Using MATLAB with PETSc](ch_matlab))
5852: . -viewer_socket_machine <machine> - Machine to use for socket
5853: . -viewer_socket_port <port> - Port number to use for socket
5854: - -mat_view binary:filename[:append] - Save matrix to file in binary format
5856: Level: beginner
5858: .seealso: [](ch_matrices), `Mat`, `MatAssemblyBegin()`, `MatSetValues()`, `PetscDrawOpenX()`, `PetscDrawCreate()`, `MatView()`, `MatAssembled()`, `PetscViewerSocketOpen()`
5859: @*/
5860: PetscErrorCode MatAssemblyEnd(Mat mat, MatAssemblyType type)
5861: {
5862: static PetscInt inassm = 0;
5863: PetscBool flg = PETSC_FALSE;
5865: PetscFunctionBegin;
5869: inassm++;
5870: MatAssemblyEnd_InUse++;
5871: if (MatAssemblyEnd_InUse == 1) { /* Do the logging only the first time through */
5872: PetscCall(PetscLogEventBegin(MAT_AssemblyEnd, mat, 0, 0, 0));
5873: PetscTryTypeMethod(mat, assemblyend, type);
5874: PetscCall(PetscLogEventEnd(MAT_AssemblyEnd, mat, 0, 0, 0));
5875: } else PetscTryTypeMethod(mat, assemblyend, type);
5877: /* Flush assembly is not a true assembly */
5878: if (type != MAT_FLUSH_ASSEMBLY) {
5879: if (mat->num_ass) {
5880: if (!mat->symmetry_eternal) {
5881: mat->symmetric = PETSC_BOOL3_UNKNOWN;
5882: mat->hermitian = PETSC_BOOL3_UNKNOWN;
5883: }
5884: if (!mat->structural_symmetry_eternal && mat->ass_nonzerostate != mat->nonzerostate) mat->structurally_symmetric = PETSC_BOOL3_UNKNOWN;
5885: if (!mat->spd_eternal) mat->spd = PETSC_BOOL3_UNKNOWN;
5886: }
5887: mat->num_ass++;
5888: mat->assembled = PETSC_TRUE;
5889: mat->ass_nonzerostate = mat->nonzerostate;
5890: }
5892: mat->insertmode = NOT_SET_VALUES;
5893: MatAssemblyEnd_InUse--;
5894: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
5895: if (inassm == 1 && type != MAT_FLUSH_ASSEMBLY) {
5896: PetscCall(MatViewFromOptions(mat, NULL, "-mat_view"));
5898: if (mat->checksymmetryonassembly) {
5899: PetscCall(MatIsSymmetric(mat, mat->checksymmetrytol, &flg));
5900: if (flg) {
5901: PetscCall(PetscPrintf(PetscObjectComm((PetscObject)mat), "Matrix is symmetric (tolerance %g)\n", (double)mat->checksymmetrytol));
5902: } else {
5903: PetscCall(PetscPrintf(PetscObjectComm((PetscObject)mat), "Matrix is not symmetric (tolerance %g)\n", (double)mat->checksymmetrytol));
5904: }
5905: }
5906: if (mat->nullsp && mat->checknullspaceonassembly) PetscCall(MatNullSpaceTest(mat->nullsp, mat, NULL));
5907: }
5908: inassm--;
5909: PetscFunctionReturn(PETSC_SUCCESS);
5910: }
5912: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
5913: /*@
5914: MatSetOption - Sets a parameter option for a matrix. Some options
5915: may be specific to certain storage formats. Some options
5916: determine how values will be inserted (or added). Sorted,
5917: row-oriented input will generally assemble the fastest. The default
5918: is row-oriented.
5920: Logically Collective for certain operations, such as `MAT_SPD`, not collective for `MAT_ROW_ORIENTED`, see `MatOption`
5922: Input Parameters:
5923: + mat - the matrix
5924: . op - the option, one of those listed below (and possibly others),
5925: - flg - turn the option on (`PETSC_TRUE`) or off (`PETSC_FALSE`)
5927: Options Describing Matrix Structure:
5928: + `MAT_SPD` - symmetric positive definite
5929: . `MAT_SYMMETRIC` - symmetric in terms of both structure and value
5930: . `MAT_HERMITIAN` - transpose is the complex conjugation
5931: . `MAT_STRUCTURALLY_SYMMETRIC` - symmetric nonzero structure
5932: . `MAT_SYMMETRY_ETERNAL` - indicates the symmetry (or Hermitian structure) or its absence will persist through any changes to the matrix
5933: . `MAT_STRUCTURAL_SYMMETRY_ETERNAL` - indicates the structural symmetry or its absence will persist through any changes to the matrix
5934: . `MAT_SPD_ETERNAL` - indicates the value of `MAT_SPD` (true or false) will persist through any changes to the matrix
5936: These are not really options of the matrix, they are knowledge about the structure of the matrix that users may provide so that they
5937: do not need to be computed (usually at a high cost)
5939: Options For Use with `MatSetValues()`:
5940: Insert a logically dense subblock, which can be
5941: . `MAT_ROW_ORIENTED` - row-oriented (default)
5943: These options reflect the data you pass in with `MatSetValues()`; it has
5944: nothing to do with how the data is stored internally in the matrix
5945: data structure.
5947: When (re)assembling a matrix, we can restrict the input for
5948: efficiency/debugging purposes. These options include
5949: . `MAT_NEW_NONZERO_LOCATIONS` - additional insertions will be allowed if they generate a new nonzero (slow)
5950: . `MAT_FORCE_DIAGONAL_ENTRIES` - forces diagonal entries to be allocated
5951: . `MAT_IGNORE_OFF_PROC_ENTRIES` - drops off-processor entries
5952: . `MAT_NEW_NONZERO_LOCATION_ERR` - generates an error for new matrix entry
5953: . `MAT_USE_HASH_TABLE` - uses a hash table to speed up matrix assembly
5954: . `MAT_NO_OFF_PROC_ENTRIES` - you know each process will only set values for its own rows, will generate an error if
5955: any process sets values for another process. This avoids all reductions in the MatAssembly routines and thus improves
5956: performance for very large process counts.
5957: - `MAT_SUBSET_OFF_PROC_ENTRIES` - you know that the first assembly after setting this flag will set a superset
5958: of the off-process entries required for all subsequent assemblies. This avoids a rendezvous step in the MatAssembly
5959: functions, instead sending only neighbor messages.
5961: Level: intermediate
5963: Notes:
5964: Except for `MAT_UNUSED_NONZERO_LOCATION_ERR` and `MAT_ROW_ORIENTED` all processes that share the matrix must pass the same value in flg!
5966: Some options are relevant only for particular matrix types and
5967: are thus ignored by others. Other options are not supported by
5968: certain matrix types and will generate an error message if set.
5970: If using Fortran to compute a matrix, one may need to
5971: use the column-oriented option (or convert to the row-oriented
5972: format).
5974: `MAT_NEW_NONZERO_LOCATIONS` set to `PETSC_FALSE` indicates that any add or insertion
5975: that would generate a new entry in the nonzero structure is instead
5976: ignored. Thus, if memory has not already been allocated for this particular
5977: data, then the insertion is ignored. For dense matrices, in which
5978: the entire array is allocated, no entries are ever ignored.
5979: Set after the first `MatAssemblyEnd()`. If this option is set, then the `MatAssemblyBegin()`/`MatAssemblyEnd()` processes has one less global reduction
5981: `MAT_NEW_NONZERO_LOCATION_ERR` set to PETSC_TRUE indicates that any add or insertion
5982: that would generate a new entry in the nonzero structure instead produces
5983: 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
5985: `MAT_NEW_NONZERO_ALLOCATION_ERR` set to `PETSC_TRUE` indicates that any add or insertion
5986: that would generate a new entry that has not been preallocated will
5987: instead produce an error. (Currently supported for `MATAIJ` and `MATBAIJ` formats
5988: only.) This is a useful flag when debugging matrix memory preallocation.
5989: If this option is set, then the `MatAssemblyBegin()`/`MatAssemblyEnd()` processes has one less global reduction
5991: `MAT_IGNORE_OFF_PROC_ENTRIES` set to `PETSC_TRUE` indicates entries destined for
5992: other processors should be dropped, rather than stashed.
5993: This is useful if you know that the "owning" processor is also
5994: always generating the correct matrix entries, so that PETSc need
5995: not transfer duplicate entries generated on another processor.
5997: `MAT_USE_HASH_TABLE` indicates that a hash table be used to improve the
5998: searches during matrix assembly. When this flag is set, the hash table
5999: is created during the first matrix assembly. This hash table is
6000: used the next time through, during `MatSetValues()`/`MatSetValuesBlocked()`
6001: to improve the searching of indices. `MAT_NEW_NONZERO_LOCATIONS` flag
6002: should be used with `MAT_USE_HASH_TABLE` flag. This option is currently
6003: supported by `MATMPIBAIJ` format only.
6005: `MAT_KEEP_NONZERO_PATTERN` indicates when `MatZeroRows()` is called the zeroed entries
6006: are kept in the nonzero structure. This flag is not used for `MatZeroRowsColumns()`
6008: `MAT_IGNORE_ZERO_ENTRIES` - for `MATAIJ` and `MATIS` matrices this will stop zero values from creating
6009: a zero location in the matrix
6011: `MAT_USE_INODES` - indicates using inode version of the code - works with `MATAIJ` matrix types
6013: `MAT_NO_OFF_PROC_ZERO_ROWS` - you know each process will only zero its own rows. This avoids all reductions in the
6014: zero row routines and thus improves performance for very large process counts.
6016: `MAT_IGNORE_LOWER_TRIANGULAR` - For `MATSBAIJ` matrices will ignore any insertions you make in the lower triangular
6017: part of the matrix (since they should match the upper triangular part).
6019: `MAT_SORTED_FULL` - each process provides exactly its local rows; all column indices for a given row are passed in a
6020: single call to `MatSetValues()`, preallocation is perfect, row-oriented, `INSERT_VALUES` is used. Common
6021: with finite difference schemes with non-periodic boundary conditions.
6023: Developer Note:
6024: `MAT_SYMMETRY_ETERNAL`, `MAT_STRUCTURAL_SYMMETRY_ETERNAL`, and `MAT_SPD_ETERNAL` are used by `MatAssemblyEnd()` and in other
6025: places where otherwise the value of `MAT_SYMMETRIC`, `MAT_STRUCTURALLY_SYMMETRIC` or `MAT_SPD` would need to be changed back
6026: to `PETSC_BOOL3_UNKNOWN` because the matrix values had changed so the code cannot be certain that the related property had
6027: not changed.
6029: .seealso: [](ch_matrices), `MatOption`, `Mat`, `MatGetOption()`
6030: @*/
6031: PetscErrorCode MatSetOption(Mat mat, MatOption op, PetscBool flg)
6032: {
6033: PetscFunctionBegin;
6035: if (op > 0) {
6038: }
6040: 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);
6042: switch (op) {
6043: case MAT_FORCE_DIAGONAL_ENTRIES:
6044: mat->force_diagonals = flg;
6045: PetscFunctionReturn(PETSC_SUCCESS);
6046: case MAT_NO_OFF_PROC_ENTRIES:
6047: mat->nooffprocentries = flg;
6048: PetscFunctionReturn(PETSC_SUCCESS);
6049: case MAT_SUBSET_OFF_PROC_ENTRIES:
6050: mat->assembly_subset = flg;
6051: if (!mat->assembly_subset) { /* See the same logic in VecAssembly wrt VEC_SUBSET_OFF_PROC_ENTRIES */
6052: #if !defined(PETSC_HAVE_MPIUNI)
6053: PetscCall(MatStashScatterDestroy_BTS(&mat->stash));
6054: #endif
6055: mat->stash.first_assembly_done = PETSC_FALSE;
6056: }
6057: PetscFunctionReturn(PETSC_SUCCESS);
6058: case MAT_NO_OFF_PROC_ZERO_ROWS:
6059: mat->nooffproczerorows = flg;
6060: PetscFunctionReturn(PETSC_SUCCESS);
6061: case MAT_SPD:
6062: if (flg) {
6063: mat->spd = PETSC_BOOL3_TRUE;
6064: mat->symmetric = PETSC_BOOL3_TRUE;
6065: mat->structurally_symmetric = PETSC_BOOL3_TRUE;
6066: } else {
6067: mat->spd = PETSC_BOOL3_FALSE;
6068: }
6069: break;
6070: case MAT_SYMMETRIC:
6071: mat->symmetric = flg ? PETSC_BOOL3_TRUE : PETSC_BOOL3_FALSE;
6072: if (flg) mat->structurally_symmetric = PETSC_BOOL3_TRUE;
6073: #if !defined(PETSC_USE_COMPLEX)
6074: mat->hermitian = flg ? PETSC_BOOL3_TRUE : PETSC_BOOL3_FALSE;
6075: #endif
6076: break;
6077: case MAT_HERMITIAN:
6078: mat->hermitian = flg ? PETSC_BOOL3_TRUE : PETSC_BOOL3_FALSE;
6079: if (flg) mat->structurally_symmetric = PETSC_BOOL3_TRUE;
6080: #if !defined(PETSC_USE_COMPLEX)
6081: mat->symmetric = flg ? PETSC_BOOL3_TRUE : PETSC_BOOL3_FALSE;
6082: #endif
6083: break;
6084: case MAT_STRUCTURALLY_SYMMETRIC:
6085: mat->structurally_symmetric = flg ? PETSC_BOOL3_TRUE : PETSC_BOOL3_FALSE;
6086: break;
6087: case MAT_SYMMETRY_ETERNAL:
6088: PetscCheck(mat->symmetric != PETSC_BOOL3_UNKNOWN, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot set MAT_SYMMETRY_ETERNAL without first setting MAT_SYMMETRIC to true or false");
6089: mat->symmetry_eternal = flg;
6090: if (flg) mat->structural_symmetry_eternal = PETSC_TRUE;
6091: break;
6092: case MAT_STRUCTURAL_SYMMETRY_ETERNAL:
6093: 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 true or false");
6094: mat->structural_symmetry_eternal = flg;
6095: break;
6096: case MAT_SPD_ETERNAL:
6097: PetscCheck(mat->spd != PETSC_BOOL3_UNKNOWN, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot set MAT_SPD_ETERNAL without first setting MAT_SPD to true or false");
6098: mat->spd_eternal = flg;
6099: if (flg) {
6100: mat->structural_symmetry_eternal = PETSC_TRUE;
6101: mat->symmetry_eternal = PETSC_TRUE;
6102: }
6103: break;
6104: case MAT_STRUCTURE_ONLY:
6105: mat->structure_only = flg;
6106: break;
6107: case MAT_SORTED_FULL:
6108: mat->sortedfull = flg;
6109: break;
6110: default:
6111: break;
6112: }
6113: PetscTryTypeMethod(mat, setoption, op, flg);
6114: PetscFunctionReturn(PETSC_SUCCESS);
6115: }
6117: /*@
6118: MatGetOption - Gets a parameter option that has been set for a matrix.
6120: Logically Collective
6122: Input Parameters:
6123: + mat - the matrix
6124: - op - the option, this only responds to certain options, check the code for which ones
6126: Output Parameter:
6127: . flg - turn the option on (`PETSC_TRUE`) or off (`PETSC_FALSE`)
6129: Level: intermediate
6131: Notes:
6132: Can only be called after `MatSetSizes()` and `MatSetType()` have been set.
6134: Certain option values may be unknown, for those use the routines `MatIsSymmetric()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, or
6135: `MatIsSymmetricKnown()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetricKnown()`
6137: .seealso: [](ch_matrices), `Mat`, `MatOption`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`,
6138: `MatIsSymmetricKnown()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetricKnown()`
6139: @*/
6140: PetscErrorCode MatGetOption(Mat mat, MatOption op, PetscBool *flg)
6141: {
6142: PetscFunctionBegin;
6146: 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);
6147: 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()");
6149: switch (op) {
6150: case MAT_NO_OFF_PROC_ENTRIES:
6151: *flg = mat->nooffprocentries;
6152: break;
6153: case MAT_NO_OFF_PROC_ZERO_ROWS:
6154: *flg = mat->nooffproczerorows;
6155: break;
6156: case MAT_SYMMETRIC:
6157: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsSymmetric() or MatIsSymmetricKnown()");
6158: break;
6159: case MAT_HERMITIAN:
6160: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsHermitian() or MatIsHermitianKnown()");
6161: break;
6162: case MAT_STRUCTURALLY_SYMMETRIC:
6163: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsStructurallySymmetric() or MatIsStructurallySymmetricKnown()");
6164: break;
6165: case MAT_SPD:
6166: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsSPDKnown()");
6167: break;
6168: case MAT_SYMMETRY_ETERNAL:
6169: *flg = mat->symmetry_eternal;
6170: break;
6171: case MAT_STRUCTURAL_SYMMETRY_ETERNAL:
6172: *flg = mat->symmetry_eternal;
6173: break;
6174: default:
6175: break;
6176: }
6177: PetscFunctionReturn(PETSC_SUCCESS);
6178: }
6180: /*@
6181: MatZeroEntries - Zeros all entries of a matrix. For sparse matrices
6182: this routine retains the old nonzero structure.
6184: Logically Collective
6186: Input Parameter:
6187: . mat - the matrix
6189: Level: intermediate
6191: Note:
6192: 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.
6193: See the Performance chapter of the users manual for information on preallocating matrices.
6195: .seealso: [](ch_matrices), `Mat`, `MatZeroRows()`, `MatZeroRowsColumns()`
6196: @*/
6197: PetscErrorCode MatZeroEntries(Mat mat)
6198: {
6199: PetscFunctionBegin;
6202: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6203: 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");
6204: MatCheckPreallocated(mat, 1);
6206: PetscCall(PetscLogEventBegin(MAT_ZeroEntries, mat, 0, 0, 0));
6207: PetscUseTypeMethod(mat, zeroentries);
6208: PetscCall(PetscLogEventEnd(MAT_ZeroEntries, mat, 0, 0, 0));
6209: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6210: PetscFunctionReturn(PETSC_SUCCESS);
6211: }
6213: /*@
6214: MatZeroRowsColumns - Zeros all entries (except possibly the main diagonal)
6215: of a set of rows and columns of a matrix.
6217: Collective
6219: Input Parameters:
6220: + mat - the matrix
6221: . numRows - the number of rows/columns to zero
6222: . rows - the global row indices
6223: . diag - value put in the diagonal of the eliminated rows
6224: . x - optional vector of the solution for zeroed rows (other entries in vector are not used), these must be set before this call
6225: - b - optional vector of the right-hand side, that will be adjusted by provided solution entries
6227: Level: intermediate
6229: Notes:
6230: This routine, along with `MatZeroRows()`, is typically used to eliminate known Dirichlet boundary conditions from a linear system.
6232: For each zeroed row, the value of the corresponding `b` is set to diag times the value of the corresponding `x`.
6233: 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
6235: If the resulting linear system is to be solved with `KSP` then one can (but does not have to) call `KSPSetInitialGuessNonzero()` to allow the
6236: Krylov method to take advantage of the known solution on the zeroed rows.
6238: For the parallel case, all processes that share the matrix (i.e.,
6239: those in the communicator used for matrix creation) MUST call this
6240: routine, regardless of whether any rows being zeroed are owned by
6241: them.
6243: Unlike `MatZeroRows()`, this ignores the `MAT_KEEP_NONZERO_PATTERN` option value set with `MatSetOption()`, it merely zeros those entries in the matrix, but never
6244: 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
6245: missing.
6247: Each processor can indicate any rows in the entire matrix to be zeroed (i.e. each process does NOT have to
6248: list only rows local to itself).
6250: The option `MAT_NO_OFF_PROC_ZERO_ROWS` does not apply to this routine.
6252: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRows()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6253: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6254: @*/
6255: PetscErrorCode MatZeroRowsColumns(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
6256: {
6257: PetscFunctionBegin;
6260: if (numRows) PetscAssertPointer(rows, 3);
6261: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6262: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6263: MatCheckPreallocated(mat, 1);
6265: PetscUseTypeMethod(mat, zerorowscolumns, numRows, rows, diag, x, b);
6266: PetscCall(MatViewFromOptions(mat, NULL, "-mat_view"));
6267: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6268: PetscFunctionReturn(PETSC_SUCCESS);
6269: }
6271: /*@
6272: MatZeroRowsColumnsIS - Zeros all entries (except possibly the main diagonal)
6273: of a set of rows and columns of a matrix.
6275: Collective
6277: Input Parameters:
6278: + mat - the matrix
6279: . is - the rows to zero
6280: . diag - value put in all diagonals of eliminated rows (0.0 will even eliminate diagonal entry)
6281: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6282: - b - optional vector of right-hand side, that will be adjusted by provided solution
6284: Level: intermediate
6286: Note:
6287: See `MatZeroRowsColumns()` for details on how this routine operates.
6289: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6290: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRows()`, `MatZeroRowsColumnsStencil()`
6291: @*/
6292: PetscErrorCode MatZeroRowsColumnsIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
6293: {
6294: PetscInt numRows;
6295: const PetscInt *rows;
6297: PetscFunctionBegin;
6302: PetscCall(ISGetLocalSize(is, &numRows));
6303: PetscCall(ISGetIndices(is, &rows));
6304: PetscCall(MatZeroRowsColumns(mat, numRows, rows, diag, x, b));
6305: PetscCall(ISRestoreIndices(is, &rows));
6306: PetscFunctionReturn(PETSC_SUCCESS);
6307: }
6309: /*@
6310: MatZeroRows - Zeros all entries (except possibly the main diagonal)
6311: of a set of rows of a matrix.
6313: Collective
6315: Input Parameters:
6316: + mat - the matrix
6317: . numRows - the number of rows to zero
6318: . rows - the global row indices
6319: . diag - value put in the diagonal of the zeroed rows
6320: . x - optional vector of solutions for zeroed rows (other entries in vector are not used), these must be set before this call
6321: - b - optional vector of right-hand side, that will be adjusted by provided solution entries
6323: Level: intermediate
6325: Notes:
6326: This routine, along with `MatZeroRowsColumns()`, is typically used to eliminate known Dirichlet boundary conditions from a linear system.
6328: For each zeroed row, the value of the corresponding `b` is set to `diag` times the value of the corresponding `x`.
6330: If the resulting linear system is to be solved with `KSP` then one can (but does not have to) call `KSPSetInitialGuessNonzero()` to allow the
6331: Krylov method to take advantage of the known solution on the zeroed rows.
6333: 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)
6334: from the matrix.
6336: Unlike `MatZeroRowsColumns()` for the `MATAIJ` and `MATBAIJ` matrix formats this removes the old nonzero structure, from the eliminated rows of the matrix
6337: but does not release memory. Because of this removal matrix-vector products with the adjusted matrix will be a bit faster. For the dense
6338: formats this does not alter the nonzero structure.
6340: If the option `MatSetOption`(mat,`MAT_KEEP_NONZERO_PATTERN`,`PETSC_TRUE`) the nonzero structure
6341: of the matrix is not changed the values are
6342: merely zeroed.
6344: The user can set a value in the diagonal entry (or for the `MATAIJ` format
6345: formats can optionally remove the main diagonal entry from the
6346: nonzero structure as well, by passing 0.0 as the final argument).
6348: For the parallel case, all processes that share the matrix (i.e.,
6349: those in the communicator used for matrix creation) MUST call this
6350: routine, regardless of whether any rows being zeroed are owned by
6351: them.
6353: Each processor can indicate any rows in the entire matrix to be zeroed (i.e. each process does NOT have to
6354: list only rows local to itself).
6356: You can call `MatSetOption`(mat,`MAT_NO_OFF_PROC_ZERO_ROWS`,`PETSC_TRUE`) if each process indicates only rows it
6357: owns that are to be zeroed. This saves a global synchronization in the implementation.
6359: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6360: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`, `PCREDISTRIBUTE`, `MAT_KEEP_NONZERO_PATTERN`
6361: @*/
6362: PetscErrorCode MatZeroRows(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
6363: {
6364: PetscFunctionBegin;
6367: if (numRows) PetscAssertPointer(rows, 3);
6368: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6369: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6370: MatCheckPreallocated(mat, 1);
6372: PetscUseTypeMethod(mat, zerorows, numRows, rows, diag, x, b);
6373: PetscCall(MatViewFromOptions(mat, NULL, "-mat_view"));
6374: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6375: PetscFunctionReturn(PETSC_SUCCESS);
6376: }
6378: /*@
6379: MatZeroRowsIS - Zeros all entries (except possibly the main diagonal)
6380: of a set of rows of a matrix indicated by an `IS`
6382: Collective
6384: Input Parameters:
6385: + mat - the matrix
6386: . is - index set, `IS`, of rows to remove (if `NULL` then no row is removed)
6387: . diag - value put in all diagonals of eliminated rows
6388: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6389: - b - optional vector of right-hand side, that will be adjusted by provided solution
6391: Level: intermediate
6393: Note:
6394: See `MatZeroRows()` for details on how this routine operates.
6396: .seealso: [](ch_matrices), `Mat`, `MatZeroRows()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6397: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`, `IS`
6398: @*/
6399: PetscErrorCode MatZeroRowsIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
6400: {
6401: PetscInt numRows = 0;
6402: const PetscInt *rows = NULL;
6404: PetscFunctionBegin;
6407: if (is) {
6409: PetscCall(ISGetLocalSize(is, &numRows));
6410: PetscCall(ISGetIndices(is, &rows));
6411: }
6412: PetscCall(MatZeroRows(mat, numRows, rows, diag, x, b));
6413: if (is) PetscCall(ISRestoreIndices(is, &rows));
6414: PetscFunctionReturn(PETSC_SUCCESS);
6415: }
6417: /*@
6418: MatZeroRowsStencil - Zeros all entries (except possibly the main diagonal)
6419: of a set of rows of a matrix indicated by a `MatStencil`. These rows must be local to the process.
6421: Collective
6423: Input Parameters:
6424: + mat - the matrix
6425: . numRows - the number of rows to remove
6426: . rows - the grid coordinates (and component number when dof > 1) for matrix rows indicated by an array of `MatStencil`
6427: . diag - value put in all diagonals of eliminated rows (0.0 will even eliminate diagonal entry)
6428: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6429: - b - optional vector of right-hand side, that will be adjusted by provided solution
6431: Level: intermediate
6433: Notes:
6434: See `MatZeroRows()` for details on how this routine operates.
6436: The grid coordinates are across the entire grid, not just the local portion
6438: For periodic boundary conditions use negative indices for values to the left (below 0; that are to be
6439: obtained by wrapping values from right edge). For values to the right of the last entry using that index plus one
6440: etc to obtain values that obtained by wrapping the values from the left edge. This does not work for anything but the
6441: `DM_BOUNDARY_PERIODIC` boundary type.
6443: 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
6444: a single value per point) you can skip filling those indices.
6446: Fortran Note:
6447: `idxm` and `idxn` should be declared as
6448: $ MatStencil idxm(4, m)
6449: and the values inserted using
6450: .vb
6451: idxm(MatStencil_i, 1) = i
6452: idxm(MatStencil_j, 1) = j
6453: idxm(MatStencil_k, 1) = k
6454: idxm(MatStencil_c, 1) = c
6455: etc
6456: .ve
6458: .seealso: [](ch_matrices), `Mat`, `MatStencil`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRows()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6459: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6460: @*/
6461: PetscErrorCode MatZeroRowsStencil(Mat mat, PetscInt numRows, const MatStencil rows[], PetscScalar diag, Vec x, Vec b)
6462: {
6463: PetscInt dim = mat->stencil.dim;
6464: PetscInt sdim = dim - (1 - (PetscInt)mat->stencil.noc);
6465: PetscInt *dims = mat->stencil.dims + 1;
6466: PetscInt *starts = mat->stencil.starts;
6467: PetscInt *dxm = (PetscInt *)rows;
6468: PetscInt *jdxm, i, j, tmp, numNewRows = 0;
6470: PetscFunctionBegin;
6473: if (numRows) PetscAssertPointer(rows, 3);
6475: PetscCall(PetscMalloc1(numRows, &jdxm));
6476: for (i = 0; i < numRows; ++i) {
6477: /* Skip unused dimensions (they are ordered k, j, i, c) */
6478: for (j = 0; j < 3 - sdim; ++j) dxm++;
6479: /* Local index in X dir */
6480: tmp = *dxm++ - starts[0];
6481: /* Loop over remaining dimensions */
6482: for (j = 0; j < dim - 1; ++j) {
6483: /* If nonlocal, set index to be negative */
6484: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = PETSC_INT_MIN;
6485: /* Update local index */
6486: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
6487: }
6488: /* Skip component slot if necessary */
6489: if (mat->stencil.noc) dxm++;
6490: /* Local row number */
6491: if (tmp >= 0) jdxm[numNewRows++] = tmp;
6492: }
6493: PetscCall(MatZeroRowsLocal(mat, numNewRows, jdxm, diag, x, b));
6494: PetscCall(PetscFree(jdxm));
6495: PetscFunctionReturn(PETSC_SUCCESS);
6496: }
6498: /*@
6499: MatZeroRowsColumnsStencil - Zeros all row and column entries (except possibly the main diagonal)
6500: of a set of rows and columns of a matrix.
6502: Collective
6504: Input Parameters:
6505: + mat - the matrix
6506: . numRows - the number of rows/columns to remove
6507: . rows - the grid coordinates (and component number when dof > 1) for matrix rows
6508: . diag - value put in all diagonals of eliminated rows (0.0 will even eliminate diagonal entry)
6509: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6510: - b - optional vector of right-hand side, that will be adjusted by provided solution
6512: Level: intermediate
6514: Notes:
6515: See `MatZeroRowsColumns()` for details on how this routine operates.
6517: The grid coordinates are across the entire grid, not just the local portion
6519: For periodic boundary conditions use negative indices for values to the left (below 0; that are to be
6520: obtained by wrapping values from right edge). For values to the right of the last entry using that index plus one
6521: etc to obtain values that obtained by wrapping the values from the left edge. This does not work for anything but the
6522: `DM_BOUNDARY_PERIODIC` boundary type.
6524: 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
6525: a single value per point) you can skip filling those indices.
6527: Fortran Note:
6528: `idxm` and `idxn` should be declared as
6529: $ MatStencil idxm(4, m)
6530: and the values inserted using
6531: .vb
6532: idxm(MatStencil_i, 1) = i
6533: idxm(MatStencil_j, 1) = j
6534: idxm(MatStencil_k, 1) = k
6535: idxm(MatStencil_c, 1) = c
6536: etc
6537: .ve
6539: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6540: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRows()`
6541: @*/
6542: PetscErrorCode MatZeroRowsColumnsStencil(Mat mat, PetscInt numRows, const MatStencil rows[], PetscScalar diag, Vec x, Vec b)
6543: {
6544: PetscInt dim = mat->stencil.dim;
6545: PetscInt sdim = dim - (1 - (PetscInt)mat->stencil.noc);
6546: PetscInt *dims = mat->stencil.dims + 1;
6547: PetscInt *starts = mat->stencil.starts;
6548: PetscInt *dxm = (PetscInt *)rows;
6549: PetscInt *jdxm, i, j, tmp, numNewRows = 0;
6551: PetscFunctionBegin;
6554: if (numRows) PetscAssertPointer(rows, 3);
6556: PetscCall(PetscMalloc1(numRows, &jdxm));
6557: for (i = 0; i < numRows; ++i) {
6558: /* Skip unused dimensions (they are ordered k, j, i, c) */
6559: for (j = 0; j < 3 - sdim; ++j) dxm++;
6560: /* Local index in X dir */
6561: tmp = *dxm++ - starts[0];
6562: /* Loop over remaining dimensions */
6563: for (j = 0; j < dim - 1; ++j) {
6564: /* If nonlocal, set index to be negative */
6565: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = PETSC_INT_MIN;
6566: /* Update local index */
6567: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
6568: }
6569: /* Skip component slot if necessary */
6570: if (mat->stencil.noc) dxm++;
6571: /* Local row number */
6572: if (tmp >= 0) jdxm[numNewRows++] = tmp;
6573: }
6574: PetscCall(MatZeroRowsColumnsLocal(mat, numNewRows, jdxm, diag, x, b));
6575: PetscCall(PetscFree(jdxm));
6576: PetscFunctionReturn(PETSC_SUCCESS);
6577: }
6579: /*@
6580: MatZeroRowsLocal - Zeros all entries (except possibly the main diagonal)
6581: of a set of rows of a matrix; using local numbering of rows.
6583: Collective
6585: Input Parameters:
6586: + mat - the matrix
6587: . numRows - the number of rows to remove
6588: . rows - the local row indices
6589: . diag - value put in all diagonals of eliminated rows
6590: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6591: - b - optional vector of right-hand side, that will be adjusted by provided solution
6593: Level: intermediate
6595: Notes:
6596: Before calling `MatZeroRowsLocal()`, the user must first set the
6597: local-to-global mapping by calling MatSetLocalToGlobalMapping(), this is often already set for matrices obtained with `DMCreateMatrix()`.
6599: See `MatZeroRows()` for details on how this routine operates.
6601: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRows()`, `MatSetOption()`,
6602: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6603: @*/
6604: PetscErrorCode MatZeroRowsLocal(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
6605: {
6606: PetscFunctionBegin;
6609: if (numRows) PetscAssertPointer(rows, 3);
6610: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6611: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6612: MatCheckPreallocated(mat, 1);
6614: if (mat->ops->zerorowslocal) {
6615: PetscUseTypeMethod(mat, zerorowslocal, numRows, rows, diag, x, b);
6616: } else {
6617: IS is, newis;
6618: const PetscInt *newRows;
6620: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Need to provide local to global mapping to matrix first");
6621: PetscCall(ISCreateGeneral(PETSC_COMM_SELF, numRows, rows, PETSC_COPY_VALUES, &is));
6622: PetscCall(ISLocalToGlobalMappingApplyIS(mat->rmap->mapping, is, &newis));
6623: PetscCall(ISGetIndices(newis, &newRows));
6624: PetscUseTypeMethod(mat, zerorows, numRows, newRows, diag, x, b);
6625: PetscCall(ISRestoreIndices(newis, &newRows));
6626: PetscCall(ISDestroy(&newis));
6627: PetscCall(ISDestroy(&is));
6628: }
6629: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6630: PetscFunctionReturn(PETSC_SUCCESS);
6631: }
6633: /*@
6634: MatZeroRowsLocalIS - Zeros all entries (except possibly the main diagonal)
6635: of a set of rows of a matrix; using local numbering of rows.
6637: Collective
6639: Input Parameters:
6640: + mat - the matrix
6641: . is - index set of rows to remove
6642: . diag - value put in all diagonals of eliminated rows
6643: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6644: - b - optional vector of right-hand side, that will be adjusted by provided solution
6646: Level: intermediate
6648: Notes:
6649: Before calling `MatZeroRowsLocalIS()`, the user must first set the
6650: local-to-global mapping by calling `MatSetLocalToGlobalMapping()`, this is often already set for matrices obtained with `DMCreateMatrix()`.
6652: See `MatZeroRows()` for details on how this routine operates.
6654: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRows()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6655: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6656: @*/
6657: PetscErrorCode MatZeroRowsLocalIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
6658: {
6659: PetscInt numRows;
6660: const PetscInt *rows;
6662: PetscFunctionBegin;
6666: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6667: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6668: MatCheckPreallocated(mat, 1);
6670: PetscCall(ISGetLocalSize(is, &numRows));
6671: PetscCall(ISGetIndices(is, &rows));
6672: PetscCall(MatZeroRowsLocal(mat, numRows, rows, diag, x, b));
6673: PetscCall(ISRestoreIndices(is, &rows));
6674: PetscFunctionReturn(PETSC_SUCCESS);
6675: }
6677: /*@
6678: MatZeroRowsColumnsLocal - Zeros all entries (except possibly the main diagonal)
6679: of a set of rows and columns of a matrix; using local numbering of rows.
6681: Collective
6683: Input Parameters:
6684: + mat - the matrix
6685: . numRows - the number of rows to remove
6686: . rows - the global row indices
6687: . diag - value put in all diagonals of eliminated rows
6688: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6689: - b - optional vector of right-hand side, that will be adjusted by provided solution
6691: Level: intermediate
6693: Notes:
6694: Before calling `MatZeroRowsColumnsLocal()`, the user must first set the
6695: local-to-global mapping by calling `MatSetLocalToGlobalMapping()`, this is often already set for matrices obtained with `DMCreateMatrix()`.
6697: See `MatZeroRowsColumns()` for details on how this routine operates.
6699: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6700: `MatZeroRows()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6701: @*/
6702: PetscErrorCode MatZeroRowsColumnsLocal(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
6703: {
6704: IS is, newis;
6705: const PetscInt *newRows;
6707: PetscFunctionBegin;
6710: if (numRows) PetscAssertPointer(rows, 3);
6711: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6712: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6713: MatCheckPreallocated(mat, 1);
6715: PetscCheck(mat->cmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Need to provide local to global mapping to matrix first");
6716: PetscCall(ISCreateGeneral(PETSC_COMM_SELF, numRows, rows, PETSC_COPY_VALUES, &is));
6717: PetscCall(ISLocalToGlobalMappingApplyIS(mat->cmap->mapping, is, &newis));
6718: PetscCall(ISGetIndices(newis, &newRows));
6719: PetscUseTypeMethod(mat, zerorowscolumns, numRows, newRows, diag, x, b);
6720: PetscCall(ISRestoreIndices(newis, &newRows));
6721: PetscCall(ISDestroy(&newis));
6722: PetscCall(ISDestroy(&is));
6723: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6724: PetscFunctionReturn(PETSC_SUCCESS);
6725: }
6727: /*@
6728: MatZeroRowsColumnsLocalIS - Zeros all entries (except possibly the main diagonal)
6729: of a set of rows and columns of a matrix; using local numbering of rows.
6731: Collective
6733: Input Parameters:
6734: + mat - the matrix
6735: . is - index set of rows to remove
6736: . diag - value put in all diagonals of eliminated rows
6737: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6738: - b - optional vector of right-hand side, that will be adjusted by provided solution
6740: Level: intermediate
6742: Notes:
6743: Before calling `MatZeroRowsColumnsLocalIS()`, the user must first set the
6744: local-to-global mapping by calling `MatSetLocalToGlobalMapping()`, this is often already set for matrices obtained with `DMCreateMatrix()`.
6746: See `MatZeroRowsColumns()` for details on how this routine operates.
6748: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6749: `MatZeroRowsColumnsLocal()`, `MatZeroRows()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6750: @*/
6751: PetscErrorCode MatZeroRowsColumnsLocalIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
6752: {
6753: PetscInt numRows;
6754: const PetscInt *rows;
6756: PetscFunctionBegin;
6760: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6761: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6762: MatCheckPreallocated(mat, 1);
6764: PetscCall(ISGetLocalSize(is, &numRows));
6765: PetscCall(ISGetIndices(is, &rows));
6766: PetscCall(MatZeroRowsColumnsLocal(mat, numRows, rows, diag, x, b));
6767: PetscCall(ISRestoreIndices(is, &rows));
6768: PetscFunctionReturn(PETSC_SUCCESS);
6769: }
6771: /*@
6772: MatGetSize - Returns the numbers of rows and columns in a matrix.
6774: Not Collective
6776: Input Parameter:
6777: . mat - the matrix
6779: Output Parameters:
6780: + m - the number of global rows
6781: - n - the number of global columns
6783: Level: beginner
6785: Note:
6786: Both output parameters can be `NULL` on input.
6788: .seealso: [](ch_matrices), `Mat`, `MatSetSizes()`, `MatGetLocalSize()`
6789: @*/
6790: PetscErrorCode MatGetSize(Mat mat, PetscInt *m, PetscInt *n)
6791: {
6792: PetscFunctionBegin;
6794: if (m) *m = mat->rmap->N;
6795: if (n) *n = mat->cmap->N;
6796: PetscFunctionReturn(PETSC_SUCCESS);
6797: }
6799: /*@
6800: MatGetLocalSize - For most matrix formats, excluding `MATELEMENTAL` and `MATSCALAPACK`, Returns the number of local rows and local columns
6801: of a matrix. For all matrices this is the local size of the left and right vectors as returned by `MatCreateVecs()`.
6803: Not Collective
6805: Input Parameter:
6806: . mat - the matrix
6808: Output Parameters:
6809: + m - the number of local rows, use `NULL` to not obtain this value
6810: - n - the number of local columns, use `NULL` to not obtain this value
6812: Level: beginner
6814: .seealso: [](ch_matrices), `Mat`, `MatSetSizes()`, `MatGetSize()`
6815: @*/
6816: PetscErrorCode MatGetLocalSize(Mat mat, PetscInt *m, PetscInt *n)
6817: {
6818: PetscFunctionBegin;
6820: if (m) PetscAssertPointer(m, 2);
6821: if (n) PetscAssertPointer(n, 3);
6822: if (m) *m = mat->rmap->n;
6823: if (n) *n = mat->cmap->n;
6824: PetscFunctionReturn(PETSC_SUCCESS);
6825: }
6827: /*@
6828: MatGetOwnershipRangeColumn - Returns the range of matrix columns associated with rows of a
6829: vector one multiplies this matrix by that are owned by this processor.
6831: Not Collective, unless matrix has not been allocated, then collective
6833: Input Parameter:
6834: . mat - the matrix
6836: Output Parameters:
6837: + m - the global index of the first local column, use `NULL` to not obtain this value
6838: - n - one more than the global index of the last local column, use `NULL` to not obtain this value
6840: Level: developer
6842: Notes:
6843: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
6845: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
6846: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
6848: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
6849: the local values in the matrix.
6851: Returns the columns of the "diagonal block" for most sparse matrix formats. See [Matrix
6852: Layouts](sec_matlayout) for details on matrix layouts.
6854: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRange()`, `MatGetOwnershipRanges()`, `MatGetOwnershipRangesColumn()`, `PetscLayout`,
6855: `MatSetSizes()`, `MatCreateAIJ()`, `DMDAGetGhostCorners()`, `DM`
6856: @*/
6857: PetscErrorCode MatGetOwnershipRangeColumn(Mat mat, PetscInt *m, PetscInt *n)
6858: {
6859: PetscFunctionBegin;
6862: if (m) PetscAssertPointer(m, 2);
6863: if (n) PetscAssertPointer(n, 3);
6864: MatCheckPreallocated(mat, 1);
6865: if (m) *m = mat->cmap->rstart;
6866: if (n) *n = mat->cmap->rend;
6867: PetscFunctionReturn(PETSC_SUCCESS);
6868: }
6870: /*@
6871: MatGetOwnershipRange - For matrices that own values by row, excludes `MATELEMENTAL` and `MATSCALAPACK`, returns the range of matrix rows owned by
6872: this MPI process.
6874: Not Collective
6876: Input Parameter:
6877: . mat - the matrix
6879: Output Parameters:
6880: + m - the global index of the first local row, use `NULL` to not obtain this value
6881: - n - one more than the global index of the last local row, use `NULL` to not obtain this value
6883: Level: beginner
6885: Notes:
6886: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
6888: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
6889: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
6891: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
6892: the local values in the matrix.
6894: The high argument is one more than the last element stored locally.
6896: For all matrices it returns the range of matrix rows associated with rows of a vector that
6897: would contain the result of a matrix vector product with this matrix. See [Matrix
6898: Layouts](sec_matlayout) for details on matrix layouts.
6900: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRanges()`, `MatGetOwnershipRangeColumn()`, `MatGetOwnershipRangesColumn()`, `PetscSplitOwnership()`,
6901: `PetscSplitOwnershipBlock()`, `PetscLayout`, `MatSetSizes()`, `MatCreateAIJ()`, `DMDAGetGhostCorners()`, `DM`
6902: @*/
6903: PetscErrorCode MatGetOwnershipRange(Mat mat, PetscInt *m, PetscInt *n)
6904: {
6905: PetscFunctionBegin;
6908: if (m) PetscAssertPointer(m, 2);
6909: if (n) PetscAssertPointer(n, 3);
6910: MatCheckPreallocated(mat, 1);
6911: if (m) *m = mat->rmap->rstart;
6912: if (n) *n = mat->rmap->rend;
6913: PetscFunctionReturn(PETSC_SUCCESS);
6914: }
6916: /*@C
6917: MatGetOwnershipRanges - For matrices that own values by row, excludes `MATELEMENTAL` and
6918: `MATSCALAPACK`, returns the range of matrix rows owned by each process.
6920: Not Collective, unless matrix has not been allocated
6922: Input Parameter:
6923: . mat - the matrix
6925: Output Parameter:
6926: . ranges - start of each processors portion plus one more than the total length at the end, of length `size` + 1
6927: where `size` is the number of MPI processes used by `mat`
6929: Level: beginner
6931: Notes:
6932: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
6934: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
6935: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
6937: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
6938: the local values in the matrix.
6940: For all matrices it returns the ranges of matrix rows associated with rows of a vector that
6941: would contain the result of a matrix vector product with this matrix. See [Matrix
6942: Layouts](sec_matlayout) for details on matrix layouts.
6944: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRange()`, `MatGetOwnershipRangeColumn()`, `MatGetOwnershipRangesColumn()`, `PetscLayout`,
6945: `PetscSplitOwnership()`, `PetscSplitOwnershipBlock()`, `MatSetSizes()`, `MatCreateAIJ()`,
6946: `DMDAGetGhostCorners()`, `DM`
6947: @*/
6948: PetscErrorCode MatGetOwnershipRanges(Mat mat, const PetscInt *ranges[])
6949: {
6950: PetscFunctionBegin;
6953: MatCheckPreallocated(mat, 1);
6954: PetscCall(PetscLayoutGetRanges(mat->rmap, ranges));
6955: PetscFunctionReturn(PETSC_SUCCESS);
6956: }
6958: /*@C
6959: MatGetOwnershipRangesColumn - Returns the ranges of matrix columns associated with rows of a
6960: vector one multiplies this vector by that are owned by each processor.
6962: Not Collective, unless matrix has not been allocated
6964: Input Parameter:
6965: . mat - the matrix
6967: Output Parameter:
6968: . ranges - start of each processors portion plus one more than the total length at the end
6970: Level: beginner
6972: Notes:
6973: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
6975: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
6976: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
6978: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
6979: the local values in the matrix.
6981: Returns the columns of the "diagonal blocks", for most sparse matrix formats. See [Matrix
6982: Layouts](sec_matlayout) for details on matrix layouts.
6984: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRange()`, `MatGetOwnershipRangeColumn()`, `MatGetOwnershipRanges()`,
6985: `PetscSplitOwnership()`, `PetscSplitOwnershipBlock()`, `PetscLayout`, `MatSetSizes()`, `MatCreateAIJ()`,
6986: `DMDAGetGhostCorners()`, `DM`
6987: @*/
6988: PetscErrorCode MatGetOwnershipRangesColumn(Mat mat, const PetscInt *ranges[])
6989: {
6990: PetscFunctionBegin;
6993: MatCheckPreallocated(mat, 1);
6994: PetscCall(PetscLayoutGetRanges(mat->cmap, ranges));
6995: PetscFunctionReturn(PETSC_SUCCESS);
6996: }
6998: /*@
6999: MatGetOwnershipIS - Get row and column ownership of a matrices' values as index sets.
7001: Not Collective
7003: Input Parameter:
7004: . A - matrix
7006: Output Parameters:
7007: + rows - rows in which this process owns elements, , use `NULL` to not obtain this value
7008: - cols - columns in which this process owns elements, use `NULL` to not obtain this value
7010: Level: intermediate
7012: Note:
7013: You should call `ISDestroy()` on the returned `IS`
7015: For most matrices, excluding `MATELEMENTAL` and `MATSCALAPACK`, this corresponds to values
7016: returned by `MatGetOwnershipRange()`, `MatGetOwnershipRangeColumn()`. For `MATELEMENTAL` and
7017: `MATSCALAPACK` the ownership is more complicated. See [Matrix Layouts](sec_matlayout) for
7018: details on matrix layouts.
7020: .seealso: [](ch_matrices), `IS`, `Mat`, `MatGetOwnershipRanges()`, `MatSetValues()`, `MATELEMENTAL`, `MATSCALAPACK`
7021: @*/
7022: PetscErrorCode MatGetOwnershipIS(Mat A, IS *rows, IS *cols)
7023: {
7024: PetscErrorCode (*f)(Mat, IS *, IS *);
7026: PetscFunctionBegin;
7029: MatCheckPreallocated(A, 1);
7030: PetscCall(PetscObjectQueryFunction((PetscObject)A, "MatGetOwnershipIS_C", &f));
7031: if (f) {
7032: PetscCall((*f)(A, rows, cols));
7033: } else { /* Create a standard row-based partition, each process is responsible for ALL columns in their row block */
7034: if (rows) PetscCall(ISCreateStride(PETSC_COMM_SELF, A->rmap->n, A->rmap->rstart, 1, rows));
7035: if (cols) PetscCall(ISCreateStride(PETSC_COMM_SELF, A->cmap->N, 0, 1, cols));
7036: }
7037: PetscFunctionReturn(PETSC_SUCCESS);
7038: }
7040: /*@
7041: MatILUFactorSymbolic - Performs symbolic ILU factorization of a matrix obtained with `MatGetFactor()`
7042: Uses levels of fill only, not drop tolerance. Use `MatLUFactorNumeric()`
7043: to complete the factorization.
7045: Collective
7047: Input Parameters:
7048: + fact - the factorized matrix obtained with `MatGetFactor()`
7049: . mat - the matrix
7050: . row - row permutation
7051: . col - column permutation
7052: - info - structure containing
7053: .vb
7054: levels - number of levels of fill.
7055: expected fill - as ratio of original fill.
7056: 1 or 0 - indicating force fill on diagonal (improves robustness for matrices
7057: missing diagonal entries)
7058: .ve
7060: Level: developer
7062: Notes:
7063: See [Matrix Factorization](sec_matfactor) for additional information.
7065: Most users should employ the `KSP` interface for linear solvers
7066: instead of working directly with matrix algebra routines such as this.
7067: See, e.g., `KSPCreate()`.
7069: Uses the definition of level of fill as in Y. Saad, {cite}`saad2003`
7071: Developer Note:
7072: The Fortran interface is not autogenerated as the
7073: interface definition cannot be generated correctly [due to `MatFactorInfo`]
7075: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`
7076: `MatGetOrdering()`, `MatFactorInfo`
7077: @*/
7078: PetscErrorCode MatILUFactorSymbolic(Mat fact, Mat mat, IS row, IS col, const MatFactorInfo *info)
7079: {
7080: PetscFunctionBegin;
7085: PetscAssertPointer(info, 5);
7086: PetscAssertPointer(fact, 1);
7087: PetscCheck(info->levels >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Levels of fill negative %" PetscInt_FMT, (PetscInt)info->levels);
7088: PetscCheck(info->fill >= 1.0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Expected fill less than 1.0 %g", (double)info->fill);
7089: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7090: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7091: MatCheckPreallocated(mat, 2);
7093: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_ILUFactorSymbolic, mat, row, col, 0));
7094: PetscUseTypeMethod(fact, ilufactorsymbolic, mat, row, col, info);
7095: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_ILUFactorSymbolic, mat, row, col, 0));
7096: PetscFunctionReturn(PETSC_SUCCESS);
7097: }
7099: /*@
7100: MatICCFactorSymbolic - Performs symbolic incomplete
7101: Cholesky factorization for a symmetric matrix. Use
7102: `MatCholeskyFactorNumeric()` to complete the factorization.
7104: Collective
7106: Input Parameters:
7107: + fact - the factorized matrix obtained with `MatGetFactor()`
7108: . mat - the matrix to be factored
7109: . perm - row and column permutation
7110: - info - structure containing
7111: .vb
7112: levels - number of levels of fill.
7113: expected fill - as ratio of original fill.
7114: .ve
7116: Level: developer
7118: Notes:
7119: Most users should employ the `KSP` interface for linear solvers
7120: instead of working directly with matrix algebra routines such as this.
7121: See, e.g., `KSPCreate()`.
7123: This uses the definition of level of fill as in Y. Saad {cite}`saad2003`
7125: Developer Note:
7126: The Fortran interface is not autogenerated as the
7127: interface definition cannot be generated correctly [due to `MatFactorInfo`]
7129: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatCholeskyFactorNumeric()`, `MatCholeskyFactor()`, `MatFactorInfo`
7130: @*/
7131: PetscErrorCode MatICCFactorSymbolic(Mat fact, Mat mat, IS perm, const MatFactorInfo *info)
7132: {
7133: PetscFunctionBegin;
7137: PetscAssertPointer(info, 4);
7138: PetscAssertPointer(fact, 1);
7139: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7140: PetscCheck(info->levels >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Levels negative %" PetscInt_FMT, (PetscInt)info->levels);
7141: PetscCheck(info->fill >= 1.0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Expected fill less than 1.0 %g", (double)info->fill);
7142: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7143: MatCheckPreallocated(mat, 2);
7145: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_ICCFactorSymbolic, mat, perm, 0, 0));
7146: PetscUseTypeMethod(fact, iccfactorsymbolic, mat, perm, info);
7147: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_ICCFactorSymbolic, mat, perm, 0, 0));
7148: PetscFunctionReturn(PETSC_SUCCESS);
7149: }
7151: /*@C
7152: MatCreateSubMatrices - Extracts several submatrices from a matrix. If submat
7153: points to an array of valid matrices, they may be reused to store the new
7154: submatrices.
7156: Collective
7158: Input Parameters:
7159: + mat - the matrix
7160: . n - the number of submatrixes to be extracted (on this processor, may be zero)
7161: . irow - index set of rows to extract
7162: . icol - index set of columns to extract
7163: - scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
7165: Output Parameter:
7166: . submat - the array of submatrices
7168: Level: advanced
7170: Notes:
7171: `MatCreateSubMatrices()` can extract ONLY sequential submatrices
7172: (from both sequential and parallel matrices). Use `MatCreateSubMatrix()`
7173: to extract a parallel submatrix.
7175: Some matrix types place restrictions on the row and column
7176: indices, such as that they be sorted or that they be equal to each other.
7178: The index sets may not have duplicate entries.
7180: When extracting submatrices from a parallel matrix, each processor can
7181: form a different submatrix by setting the rows and columns of its
7182: individual index sets according to the local submatrix desired.
7184: When finished using the submatrices, the user should destroy
7185: them with `MatDestroySubMatrices()`.
7187: `MAT_REUSE_MATRIX` can only be used when the nonzero structure of the
7188: original matrix has not changed from that last call to `MatCreateSubMatrices()`.
7190: This routine creates the matrices in submat; you should NOT create them before
7191: calling it. It also allocates the array of matrix pointers submat.
7193: For `MATBAIJ` matrices the index sets must respect the block structure, that is if they
7194: request one row/column in a block, they must request all rows/columns that are in
7195: that block. For example, if the block size is 2 you cannot request just row 0 and
7196: column 0.
7198: Fortran Note:
7199: One must pass in as `submat` a `Mat` array of size at least `n`+1.
7201: .seealso: [](ch_matrices), `Mat`, `MatDestroySubMatrices()`, `MatCreateSubMatrix()`, `MatGetRow()`, `MatGetDiagonal()`, `MatReuse`
7202: @*/
7203: PetscErrorCode MatCreateSubMatrices(Mat mat, PetscInt n, const IS irow[], const IS icol[], MatReuse scall, Mat *submat[])
7204: {
7205: PetscInt i;
7206: PetscBool eq;
7208: PetscFunctionBegin;
7211: if (n) {
7212: PetscAssertPointer(irow, 3);
7214: PetscAssertPointer(icol, 4);
7216: }
7217: PetscAssertPointer(submat, 6);
7218: if (n && scall == MAT_REUSE_MATRIX) {
7219: PetscAssertPointer(*submat, 6);
7221: }
7222: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7223: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7224: MatCheckPreallocated(mat, 1);
7225: PetscCall(PetscLogEventBegin(MAT_CreateSubMats, mat, 0, 0, 0));
7226: PetscUseTypeMethod(mat, createsubmatrices, n, irow, icol, scall, submat);
7227: PetscCall(PetscLogEventEnd(MAT_CreateSubMats, mat, 0, 0, 0));
7228: for (i = 0; i < n; i++) {
7229: (*submat)[i]->factortype = MAT_FACTOR_NONE; /* in case in place factorization was previously done on submatrix */
7230: PetscCall(ISEqualUnsorted(irow[i], icol[i], &eq));
7231: if (eq) PetscCall(MatPropagateSymmetryOptions(mat, (*submat)[i]));
7232: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_HIP)
7233: if (mat->boundtocpu && mat->bindingpropagates) {
7234: PetscCall(MatBindToCPU((*submat)[i], PETSC_TRUE));
7235: PetscCall(MatSetBindingPropagates((*submat)[i], PETSC_TRUE));
7236: }
7237: #endif
7238: }
7239: PetscFunctionReturn(PETSC_SUCCESS);
7240: }
7242: /*@C
7243: MatCreateSubMatricesMPI - Extracts MPI submatrices across a sub communicator of mat (by pairs of `IS` that may live on subcomms).
7245: Collective
7247: Input Parameters:
7248: + mat - the matrix
7249: . n - the number of submatrixes to be extracted
7250: . irow - index set of rows to extract
7251: . icol - index set of columns to extract
7252: - scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
7254: Output Parameter:
7255: . submat - the array of submatrices
7257: Level: advanced
7259: Note:
7260: This is used by `PCGASM`
7262: .seealso: [](ch_matrices), `Mat`, `PCGASM`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRow()`, `MatGetDiagonal()`, `MatReuse`
7263: @*/
7264: PetscErrorCode MatCreateSubMatricesMPI(Mat mat, PetscInt n, const IS irow[], const IS icol[], MatReuse scall, Mat *submat[])
7265: {
7266: PetscInt i;
7267: PetscBool eq;
7269: PetscFunctionBegin;
7272: if (n) {
7273: PetscAssertPointer(irow, 3);
7275: PetscAssertPointer(icol, 4);
7277: }
7278: PetscAssertPointer(submat, 6);
7279: if (n && scall == MAT_REUSE_MATRIX) {
7280: PetscAssertPointer(*submat, 6);
7282: }
7283: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7284: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7285: MatCheckPreallocated(mat, 1);
7287: PetscCall(PetscLogEventBegin(MAT_CreateSubMats, mat, 0, 0, 0));
7288: PetscUseTypeMethod(mat, createsubmatricesmpi, n, irow, icol, scall, submat);
7289: PetscCall(PetscLogEventEnd(MAT_CreateSubMats, mat, 0, 0, 0));
7290: for (i = 0; i < n; i++) {
7291: PetscCall(ISEqualUnsorted(irow[i], icol[i], &eq));
7292: if (eq) PetscCall(MatPropagateSymmetryOptions(mat, (*submat)[i]));
7293: }
7294: PetscFunctionReturn(PETSC_SUCCESS);
7295: }
7297: /*@C
7298: MatDestroyMatrices - Destroys an array of matrices.
7300: Collective
7302: Input Parameters:
7303: + n - the number of local matrices
7304: - mat - the matrices (this is a pointer to the array of matrices)
7306: Level: advanced
7308: Notes:
7309: Frees not only the matrices, but also the array that contains the matrices
7311: For matrices obtained with `MatCreateSubMatrices()` use `MatDestroySubMatrices()`
7313: Fortran Note:
7314: Does not free the `mat` array.
7316: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatDestroySubMatrices()`
7317: @*/
7318: PetscErrorCode MatDestroyMatrices(PetscInt n, Mat *mat[])
7319: {
7320: PetscInt i;
7322: PetscFunctionBegin;
7323: if (!*mat) PetscFunctionReturn(PETSC_SUCCESS);
7324: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Trying to destroy negative number of matrices %" PetscInt_FMT, n);
7325: PetscAssertPointer(mat, 2);
7327: for (i = 0; i < n; i++) PetscCall(MatDestroy(&(*mat)[i]));
7329: /* memory is allocated even if n = 0 */
7330: PetscCall(PetscFree(*mat));
7331: PetscFunctionReturn(PETSC_SUCCESS);
7332: }
7334: /*@C
7335: MatDestroySubMatrices - Destroys a set of matrices obtained with `MatCreateSubMatrices()`.
7337: Collective
7339: Input Parameters:
7340: + n - the number of local matrices
7341: - mat - the matrices (this is a pointer to the array of matrices, just to match the calling
7342: sequence of `MatCreateSubMatrices()`)
7344: Level: advanced
7346: Note:
7347: Frees not only the matrices, but also the array that contains the matrices
7349: Fortran Note:
7350: Does not free the `mat` array.
7352: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatDestroyMatrices()`
7353: @*/
7354: PetscErrorCode MatDestroySubMatrices(PetscInt n, Mat *mat[])
7355: {
7356: Mat mat0;
7358: PetscFunctionBegin;
7359: if (!*mat) PetscFunctionReturn(PETSC_SUCCESS);
7360: /* mat[] is an array of length n+1, see MatCreateSubMatrices_xxx() */
7361: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Trying to destroy negative number of matrices %" PetscInt_FMT, n);
7362: PetscAssertPointer(mat, 2);
7364: mat0 = (*mat)[0];
7365: if (mat0 && mat0->ops->destroysubmatrices) {
7366: PetscCall((*mat0->ops->destroysubmatrices)(n, mat));
7367: } else {
7368: PetscCall(MatDestroyMatrices(n, mat));
7369: }
7370: PetscFunctionReturn(PETSC_SUCCESS);
7371: }
7373: /*@
7374: MatGetSeqNonzeroStructure - Extracts the nonzero structure from a matrix and stores it, in its entirety, on each process
7376: Collective
7378: Input Parameter:
7379: . mat - the matrix
7381: Output Parameter:
7382: . matstruct - the sequential matrix with the nonzero structure of `mat`
7384: Level: developer
7386: .seealso: [](ch_matrices), `Mat`, `MatDestroySeqNonzeroStructure()`, `MatCreateSubMatrices()`, `MatDestroyMatrices()`
7387: @*/
7388: PetscErrorCode MatGetSeqNonzeroStructure(Mat mat, Mat *matstruct)
7389: {
7390: PetscFunctionBegin;
7392: PetscAssertPointer(matstruct, 2);
7395: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7396: MatCheckPreallocated(mat, 1);
7398: PetscCall(PetscLogEventBegin(MAT_GetSeqNonzeroStructure, mat, 0, 0, 0));
7399: PetscUseTypeMethod(mat, getseqnonzerostructure, matstruct);
7400: PetscCall(PetscLogEventEnd(MAT_GetSeqNonzeroStructure, mat, 0, 0, 0));
7401: PetscFunctionReturn(PETSC_SUCCESS);
7402: }
7404: /*@C
7405: MatDestroySeqNonzeroStructure - Destroys matrix obtained with `MatGetSeqNonzeroStructure()`.
7407: Collective
7409: Input Parameter:
7410: . mat - the matrix
7412: Level: advanced
7414: Note:
7415: This is not needed, one can just call `MatDestroy()`
7417: .seealso: [](ch_matrices), `Mat`, `MatGetSeqNonzeroStructure()`
7418: @*/
7419: PetscErrorCode MatDestroySeqNonzeroStructure(Mat *mat)
7420: {
7421: PetscFunctionBegin;
7422: PetscAssertPointer(mat, 1);
7423: PetscCall(MatDestroy(mat));
7424: PetscFunctionReturn(PETSC_SUCCESS);
7425: }
7427: /*@
7428: MatIncreaseOverlap - Given a set of submatrices indicated by index sets,
7429: replaces the index sets by larger ones that represent submatrices with
7430: additional overlap.
7432: Collective
7434: Input Parameters:
7435: + mat - the matrix
7436: . n - the number of index sets
7437: . is - the array of index sets (these index sets will changed during the call)
7438: - ov - the additional overlap requested
7440: Options Database Key:
7441: . -mat_increase_overlap_scalable - use a scalable algorithm to compute the overlap (supported by MPIAIJ matrix)
7443: Level: developer
7445: Note:
7446: The computed overlap preserves the matrix block sizes when the blocks are square.
7447: That is: if a matrix nonzero for a given block would increase the overlap all columns associated with
7448: that block are included in the overlap regardless of whether each specific column would increase the overlap.
7450: .seealso: [](ch_matrices), `Mat`, `PCASM`, `MatSetBlockSize()`, `MatIncreaseOverlapSplit()`, `MatCreateSubMatrices()`
7451: @*/
7452: PetscErrorCode MatIncreaseOverlap(Mat mat, PetscInt n, IS is[], PetscInt ov)
7453: {
7454: PetscInt i, bs, cbs;
7456: PetscFunctionBegin;
7460: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must have one or more domains, you have %" PetscInt_FMT, n);
7461: if (n) {
7462: PetscAssertPointer(is, 3);
7464: }
7465: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7466: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7467: MatCheckPreallocated(mat, 1);
7469: if (!ov || !n) PetscFunctionReturn(PETSC_SUCCESS);
7470: PetscCall(PetscLogEventBegin(MAT_IncreaseOverlap, mat, 0, 0, 0));
7471: PetscUseTypeMethod(mat, increaseoverlap, n, is, ov);
7472: PetscCall(PetscLogEventEnd(MAT_IncreaseOverlap, mat, 0, 0, 0));
7473: PetscCall(MatGetBlockSizes(mat, &bs, &cbs));
7474: if (bs == cbs) {
7475: for (i = 0; i < n; i++) PetscCall(ISSetBlockSize(is[i], bs));
7476: }
7477: PetscFunctionReturn(PETSC_SUCCESS);
7478: }
7480: PetscErrorCode MatIncreaseOverlapSplit_Single(Mat, IS *, PetscInt);
7482: /*@
7483: MatIncreaseOverlapSplit - Given a set of submatrices indicated by index sets across
7484: a sub communicator, replaces the index sets by larger ones that represent submatrices with
7485: additional overlap.
7487: Collective
7489: Input Parameters:
7490: + mat - the matrix
7491: . n - the number of index sets
7492: . is - the array of index sets (these index sets will changed during the call)
7493: - ov - the additional overlap requested
7495: ` Options Database Key:
7496: . -mat_increase_overlap_scalable - use a scalable algorithm to compute the overlap (supported by MPIAIJ matrix)
7498: Level: developer
7500: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatIncreaseOverlap()`
7501: @*/
7502: PetscErrorCode MatIncreaseOverlapSplit(Mat mat, PetscInt n, IS is[], PetscInt ov)
7503: {
7504: PetscInt i;
7506: PetscFunctionBegin;
7509: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must have one or more domains, you have %" PetscInt_FMT, n);
7510: if (n) {
7511: PetscAssertPointer(is, 3);
7513: }
7514: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7515: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7516: MatCheckPreallocated(mat, 1);
7517: if (!ov) PetscFunctionReturn(PETSC_SUCCESS);
7518: PetscCall(PetscLogEventBegin(MAT_IncreaseOverlap, mat, 0, 0, 0));
7519: for (i = 0; i < n; i++) PetscCall(MatIncreaseOverlapSplit_Single(mat, &is[i], ov));
7520: PetscCall(PetscLogEventEnd(MAT_IncreaseOverlap, mat, 0, 0, 0));
7521: PetscFunctionReturn(PETSC_SUCCESS);
7522: }
7524: /*@
7525: MatGetBlockSize - Returns the matrix block size.
7527: Not Collective
7529: Input Parameter:
7530: . mat - the matrix
7532: Output Parameter:
7533: . bs - block size
7535: Level: intermediate
7537: Notes:
7538: Block row formats are `MATBAIJ` and `MATSBAIJ` ALWAYS have square block storage in the matrix.
7540: If the block size has not been set yet this routine returns 1.
7542: .seealso: [](ch_matrices), `Mat`, `MATBAIJ`, `MATSBAIJ`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSizes()`
7543: @*/
7544: PetscErrorCode MatGetBlockSize(Mat mat, PetscInt *bs)
7545: {
7546: PetscFunctionBegin;
7548: PetscAssertPointer(bs, 2);
7549: *bs = PetscAbs(mat->rmap->bs);
7550: PetscFunctionReturn(PETSC_SUCCESS);
7551: }
7553: /*@
7554: MatGetBlockSizes - Returns the matrix block row and column sizes.
7556: Not Collective
7558: Input Parameter:
7559: . mat - the matrix
7561: Output Parameters:
7562: + rbs - row block size
7563: - cbs - column block size
7565: Level: intermediate
7567: Notes:
7568: Block row formats are `MATBAIJ` and `MATSBAIJ` ALWAYS have square block storage in the matrix.
7569: If you pass a different block size for the columns than the rows, the row block size determines the square block storage.
7571: If a block size has not been set yet this routine returns 1.
7573: .seealso: [](ch_matrices), `Mat`, `MATBAIJ`, `MATSBAIJ`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSize()`, `MatSetBlockSizes()`
7574: @*/
7575: PetscErrorCode MatGetBlockSizes(Mat mat, PetscInt *rbs, PetscInt *cbs)
7576: {
7577: PetscFunctionBegin;
7579: if (rbs) PetscAssertPointer(rbs, 2);
7580: if (cbs) PetscAssertPointer(cbs, 3);
7581: if (rbs) *rbs = PetscAbs(mat->rmap->bs);
7582: if (cbs) *cbs = PetscAbs(mat->cmap->bs);
7583: PetscFunctionReturn(PETSC_SUCCESS);
7584: }
7586: /*@
7587: MatSetBlockSize - Sets the matrix block size.
7589: Logically Collective
7591: Input Parameters:
7592: + mat - the matrix
7593: - bs - block size
7595: Level: intermediate
7597: Notes:
7598: Block row formats are `MATBAIJ` and `MATSBAIJ` formats ALWAYS have square block storage in the matrix.
7599: This must be called before `MatSetUp()` or MatXXXSetPreallocation() (or will default to 1) and the block size cannot be changed later.
7601: For `MATAIJ` matrix format, this function can be called at a later stage, provided that the specified block size
7602: is compatible with the matrix local sizes.
7604: .seealso: [](ch_matrices), `Mat`, `MATBAIJ`, `MATSBAIJ`, `MATAIJ`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`, `MatGetBlockSizes()`
7605: @*/
7606: PetscErrorCode MatSetBlockSize(Mat mat, PetscInt bs)
7607: {
7608: PetscFunctionBegin;
7611: PetscCall(MatSetBlockSizes(mat, bs, bs));
7612: PetscFunctionReturn(PETSC_SUCCESS);
7613: }
7615: typedef struct {
7616: PetscInt n;
7617: IS *is;
7618: Mat *mat;
7619: PetscObjectState nonzerostate;
7620: Mat C;
7621: } EnvelopeData;
7623: static PetscErrorCode EnvelopeDataDestroy(void *ptr)
7624: {
7625: EnvelopeData *edata = (EnvelopeData *)ptr;
7627: PetscFunctionBegin;
7628: for (PetscInt i = 0; i < edata->n; i++) PetscCall(ISDestroy(&edata->is[i]));
7629: PetscCall(PetscFree(edata->is));
7630: PetscCall(PetscFree(edata));
7631: PetscFunctionReturn(PETSC_SUCCESS);
7632: }
7634: /*@
7635: MatComputeVariableBlockEnvelope - Given a matrix whose nonzeros are in blocks along the diagonal this computes and stores
7636: the sizes of these blocks in the matrix. An individual block may lie over several processes.
7638: Collective
7640: Input Parameter:
7641: . mat - the matrix
7643: Level: intermediate
7645: Notes:
7646: There can be zeros within the blocks
7648: The blocks can overlap between processes, including laying on more than two processes
7650: .seealso: [](ch_matrices), `Mat`, `MatInvertVariableBlockEnvelope()`, `MatSetVariableBlockSizes()`
7651: @*/
7652: PetscErrorCode MatComputeVariableBlockEnvelope(Mat mat)
7653: {
7654: PetscInt n, *sizes, *starts, i = 0, env = 0, tbs = 0, lblocks = 0, rstart, II, ln = 0, cnt = 0, cstart, cend;
7655: PetscInt *diag, *odiag, sc;
7656: VecScatter scatter;
7657: PetscScalar *seqv;
7658: const PetscScalar *parv;
7659: const PetscInt *ia, *ja;
7660: PetscBool set, flag, done;
7661: Mat AA = mat, A;
7662: MPI_Comm comm;
7663: PetscMPIInt rank, size, tag;
7664: MPI_Status status;
7665: PetscContainer container;
7666: EnvelopeData *edata;
7667: Vec seq, par;
7668: IS isglobal;
7670: PetscFunctionBegin;
7672: PetscCall(MatIsSymmetricKnown(mat, &set, &flag));
7673: if (!set || !flag) {
7674: /* TODO: only needs nonzero structure of transpose */
7675: PetscCall(MatTranspose(mat, MAT_INITIAL_MATRIX, &AA));
7676: PetscCall(MatAXPY(AA, 1.0, mat, DIFFERENT_NONZERO_PATTERN));
7677: }
7678: PetscCall(MatAIJGetLocalMat(AA, &A));
7679: PetscCall(MatGetRowIJ(A, 0, PETSC_FALSE, PETSC_FALSE, &n, &ia, &ja, &done));
7680: PetscCheck(done, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Unable to get IJ structure from matrix");
7682: PetscCall(MatGetLocalSize(mat, &n, NULL));
7683: PetscCall(PetscObjectGetNewTag((PetscObject)mat, &tag));
7684: PetscCall(PetscObjectGetComm((PetscObject)mat, &comm));
7685: PetscCallMPI(MPI_Comm_size(comm, &size));
7686: PetscCallMPI(MPI_Comm_rank(comm, &rank));
7688: PetscCall(PetscMalloc2(n, &sizes, n, &starts));
7690: if (rank > 0) {
7691: PetscCallMPI(MPI_Recv(&env, 1, MPIU_INT, rank - 1, tag, comm, &status));
7692: PetscCallMPI(MPI_Recv(&tbs, 1, MPIU_INT, rank - 1, tag, comm, &status));
7693: }
7694: PetscCall(MatGetOwnershipRange(mat, &rstart, NULL));
7695: for (i = 0; i < n; i++) {
7696: env = PetscMax(env, ja[ia[i + 1] - 1]);
7697: II = rstart + i;
7698: if (env == II) {
7699: starts[lblocks] = tbs;
7700: sizes[lblocks++] = 1 + II - tbs;
7701: tbs = 1 + II;
7702: }
7703: }
7704: if (rank < size - 1) {
7705: PetscCallMPI(MPI_Send(&env, 1, MPIU_INT, rank + 1, tag, comm));
7706: PetscCallMPI(MPI_Send(&tbs, 1, MPIU_INT, rank + 1, tag, comm));
7707: }
7709: PetscCall(MatRestoreRowIJ(A, 0, PETSC_FALSE, PETSC_FALSE, &n, &ia, &ja, &done));
7710: if (!set || !flag) PetscCall(MatDestroy(&AA));
7711: PetscCall(MatDestroy(&A));
7713: PetscCall(PetscNew(&edata));
7714: PetscCall(MatGetNonzeroState(mat, &edata->nonzerostate));
7715: edata->n = lblocks;
7716: /* create IS needed for extracting blocks from the original matrix */
7717: PetscCall(PetscMalloc1(lblocks, &edata->is));
7718: for (PetscInt i = 0; i < lblocks; i++) PetscCall(ISCreateStride(PETSC_COMM_SELF, sizes[i], starts[i], 1, &edata->is[i]));
7720: /* Create the resulting inverse matrix nonzero structure with preallocation information */
7721: PetscCall(MatCreate(PetscObjectComm((PetscObject)mat), &edata->C));
7722: PetscCall(MatSetSizes(edata->C, mat->rmap->n, mat->cmap->n, mat->rmap->N, mat->cmap->N));
7723: PetscCall(MatSetBlockSizesFromMats(edata->C, mat, mat));
7724: PetscCall(MatSetType(edata->C, MATAIJ));
7726: /* Communicate the start and end of each row, from each block to the correct rank */
7727: /* TODO: Use PetscSF instead of VecScatter */
7728: for (PetscInt i = 0; i < lblocks; i++) ln += sizes[i];
7729: PetscCall(VecCreateSeq(PETSC_COMM_SELF, 2 * ln, &seq));
7730: PetscCall(VecGetArrayWrite(seq, &seqv));
7731: for (PetscInt i = 0; i < lblocks; i++) {
7732: for (PetscInt j = 0; j < sizes[i]; j++) {
7733: seqv[cnt] = starts[i];
7734: seqv[cnt + 1] = starts[i] + sizes[i];
7735: cnt += 2;
7736: }
7737: }
7738: PetscCall(VecRestoreArrayWrite(seq, &seqv));
7739: PetscCallMPI(MPI_Scan(&cnt, &sc, 1, MPIU_INT, MPI_SUM, PetscObjectComm((PetscObject)mat)));
7740: sc -= cnt;
7741: PetscCall(VecCreateMPI(PetscObjectComm((PetscObject)mat), 2 * mat->rmap->n, 2 * mat->rmap->N, &par));
7742: PetscCall(ISCreateStride(PETSC_COMM_SELF, cnt, sc, 1, &isglobal));
7743: PetscCall(VecScatterCreate(seq, NULL, par, isglobal, &scatter));
7744: PetscCall(ISDestroy(&isglobal));
7745: PetscCall(VecScatterBegin(scatter, seq, par, INSERT_VALUES, SCATTER_FORWARD));
7746: PetscCall(VecScatterEnd(scatter, seq, par, INSERT_VALUES, SCATTER_FORWARD));
7747: PetscCall(VecScatterDestroy(&scatter));
7748: PetscCall(VecDestroy(&seq));
7749: PetscCall(MatGetOwnershipRangeColumn(mat, &cstart, &cend));
7750: PetscCall(PetscMalloc2(mat->rmap->n, &diag, mat->rmap->n, &odiag));
7751: PetscCall(VecGetArrayRead(par, &parv));
7752: cnt = 0;
7753: PetscCall(MatGetSize(mat, NULL, &n));
7754: for (PetscInt i = 0; i < mat->rmap->n; i++) {
7755: PetscInt start, end, d = 0, od = 0;
7757: start = (PetscInt)PetscRealPart(parv[cnt]);
7758: end = (PetscInt)PetscRealPart(parv[cnt + 1]);
7759: cnt += 2;
7761: if (start < cstart) {
7762: od += cstart - start + n - cend;
7763: d += cend - cstart;
7764: } else if (start < cend) {
7765: od += n - cend;
7766: d += cend - start;
7767: } else od += n - start;
7768: if (end <= cstart) {
7769: od -= cstart - end + n - cend;
7770: d -= cend - cstart;
7771: } else if (end < cend) {
7772: od -= n - cend;
7773: d -= cend - end;
7774: } else od -= n - end;
7776: odiag[i] = od;
7777: diag[i] = d;
7778: }
7779: PetscCall(VecRestoreArrayRead(par, &parv));
7780: PetscCall(VecDestroy(&par));
7781: PetscCall(MatXAIJSetPreallocation(edata->C, mat->rmap->bs, diag, odiag, NULL, NULL));
7782: PetscCall(PetscFree2(diag, odiag));
7783: PetscCall(PetscFree2(sizes, starts));
7785: PetscCall(PetscContainerCreate(PETSC_COMM_SELF, &container));
7786: PetscCall(PetscContainerSetPointer(container, edata));
7787: PetscCall(PetscContainerSetUserDestroy(container, (PetscErrorCode (*)(void *))EnvelopeDataDestroy));
7788: PetscCall(PetscObjectCompose((PetscObject)mat, "EnvelopeData", (PetscObject)container));
7789: PetscCall(PetscObjectDereference((PetscObject)container));
7790: PetscFunctionReturn(PETSC_SUCCESS);
7791: }
7793: /*@
7794: MatInvertVariableBlockEnvelope - set matrix C to be the inverted block diagonal of matrix A
7796: Collective
7798: Input Parameters:
7799: + A - the matrix
7800: - reuse - indicates if the `C` matrix was obtained from a previous call to this routine
7802: Output Parameter:
7803: . C - matrix with inverted block diagonal of `A`
7805: Level: advanced
7807: Note:
7808: For efficiency the matrix `A` should have all the nonzero entries clustered in smallish blocks along the diagonal.
7810: .seealso: [](ch_matrices), `Mat`, `MatInvertBlockDiagonal()`, `MatComputeBlockDiagonal()`
7811: @*/
7812: PetscErrorCode MatInvertVariableBlockEnvelope(Mat A, MatReuse reuse, Mat *C)
7813: {
7814: PetscContainer container;
7815: EnvelopeData *edata;
7816: PetscObjectState nonzerostate;
7818: PetscFunctionBegin;
7819: PetscCall(PetscObjectQuery((PetscObject)A, "EnvelopeData", (PetscObject *)&container));
7820: if (!container) {
7821: PetscCall(MatComputeVariableBlockEnvelope(A));
7822: PetscCall(PetscObjectQuery((PetscObject)A, "EnvelopeData", (PetscObject *)&container));
7823: }
7824: PetscCall(PetscContainerGetPointer(container, (void **)&edata));
7825: PetscCall(MatGetNonzeroState(A, &nonzerostate));
7826: PetscCheck(nonzerostate <= edata->nonzerostate, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Cannot handle changes to matrix nonzero structure");
7827: PetscCheck(reuse != MAT_REUSE_MATRIX || *C == edata->C, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "C matrix must be the same as previously output");
7829: PetscCall(MatCreateSubMatrices(A, edata->n, edata->is, edata->is, MAT_INITIAL_MATRIX, &edata->mat));
7830: *C = edata->C;
7832: for (PetscInt i = 0; i < edata->n; i++) {
7833: Mat D;
7834: PetscScalar *dvalues;
7836: PetscCall(MatConvert(edata->mat[i], MATSEQDENSE, MAT_INITIAL_MATRIX, &D));
7837: PetscCall(MatSetOption(*C, MAT_ROW_ORIENTED, PETSC_FALSE));
7838: PetscCall(MatSeqDenseInvert(D));
7839: PetscCall(MatDenseGetArray(D, &dvalues));
7840: PetscCall(MatSetValuesIS(*C, edata->is[i], edata->is[i], dvalues, INSERT_VALUES));
7841: PetscCall(MatDestroy(&D));
7842: }
7843: PetscCall(MatDestroySubMatrices(edata->n, &edata->mat));
7844: PetscCall(MatAssemblyBegin(*C, MAT_FINAL_ASSEMBLY));
7845: PetscCall(MatAssemblyEnd(*C, MAT_FINAL_ASSEMBLY));
7846: PetscFunctionReturn(PETSC_SUCCESS);
7847: }
7849: /*@
7850: MatSetVariableBlockSizes - Sets diagonal point-blocks of the matrix that need not be of the same size
7852: Not Collective
7854: Input Parameters:
7855: + mat - the matrix
7856: . nblocks - the number of blocks on this process, each block can only exist on a single process
7857: - bsizes - the block sizes
7859: Level: intermediate
7861: Notes:
7862: Currently used by `PCVPBJACOBI` for `MATAIJ` matrices
7864: Each variable point-block set of degrees of freedom must live on a single MPI process. That is a point block cannot straddle two MPI processes.
7866: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`, `MatGetBlockSizes()`, `MatGetVariableBlockSizes()`,
7867: `MatComputeVariableBlockEnvelope()`, `PCVPBJACOBI`
7868: @*/
7869: PetscErrorCode MatSetVariableBlockSizes(Mat mat, PetscInt nblocks, const PetscInt bsizes[])
7870: {
7871: PetscInt ncnt = 0, nlocal;
7873: PetscFunctionBegin;
7875: PetscCall(MatGetLocalSize(mat, &nlocal, NULL));
7876: 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);
7877: for (PetscInt i = 0; i < nblocks; i++) ncnt += bsizes[i];
7878: 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);
7879: PetscCall(PetscFree(mat->bsizes));
7880: mat->nblocks = nblocks;
7881: PetscCall(PetscMalloc1(nblocks, &mat->bsizes));
7882: PetscCall(PetscArraycpy(mat->bsizes, bsizes, nblocks));
7883: PetscFunctionReturn(PETSC_SUCCESS);
7884: }
7886: /*@C
7887: MatGetVariableBlockSizes - Gets a diagonal blocks of the matrix that need not be of the same size
7889: Not Collective; No Fortran Support
7891: Input Parameter:
7892: . mat - the matrix
7894: Output Parameters:
7895: + nblocks - the number of blocks on this process
7896: - bsizes - the block sizes
7898: Level: intermediate
7900: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`, `MatGetBlockSizes()`, `MatSetVariableBlockSizes()`, `MatComputeVariableBlockEnvelope()`
7901: @*/
7902: PetscErrorCode MatGetVariableBlockSizes(Mat mat, PetscInt *nblocks, const PetscInt *bsizes[])
7903: {
7904: PetscFunctionBegin;
7906: if (nblocks) *nblocks = mat->nblocks;
7907: if (bsizes) *bsizes = mat->bsizes;
7908: PetscFunctionReturn(PETSC_SUCCESS);
7909: }
7911: /*@
7912: MatSetBlockSizes - Sets the matrix block row and column sizes.
7914: Logically Collective
7916: Input Parameters:
7917: + mat - the matrix
7918: . rbs - row block size
7919: - cbs - column block size
7921: Level: intermediate
7923: Notes:
7924: Block row formats are `MATBAIJ` and `MATSBAIJ`. These formats ALWAYS have square block storage in the matrix.
7925: If you pass a different block size for the columns than the rows, the row block size determines the square block storage.
7926: This must be called before `MatSetUp()` or MatXXXSetPreallocation() (or will default to 1) and the block size cannot be changed later.
7928: For `MATAIJ` matrix this function can be called at a later stage, provided that the specified block sizes
7929: are compatible with the matrix local sizes.
7931: The row and column block size determine the blocksize of the "row" and "column" vectors returned by `MatCreateVecs()`.
7933: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSize()`, `MatGetBlockSizes()`
7934: @*/
7935: PetscErrorCode MatSetBlockSizes(Mat mat, PetscInt rbs, PetscInt cbs)
7936: {
7937: PetscFunctionBegin;
7941: PetscTryTypeMethod(mat, setblocksizes, rbs, cbs);
7942: if (mat->rmap->refcnt) {
7943: ISLocalToGlobalMapping l2g = NULL;
7944: PetscLayout nmap = NULL;
7946: PetscCall(PetscLayoutDuplicate(mat->rmap, &nmap));
7947: if (mat->rmap->mapping) PetscCall(ISLocalToGlobalMappingDuplicate(mat->rmap->mapping, &l2g));
7948: PetscCall(PetscLayoutDestroy(&mat->rmap));
7949: mat->rmap = nmap;
7950: mat->rmap->mapping = l2g;
7951: }
7952: if (mat->cmap->refcnt) {
7953: ISLocalToGlobalMapping l2g = NULL;
7954: PetscLayout nmap = NULL;
7956: PetscCall(PetscLayoutDuplicate(mat->cmap, &nmap));
7957: if (mat->cmap->mapping) PetscCall(ISLocalToGlobalMappingDuplicate(mat->cmap->mapping, &l2g));
7958: PetscCall(PetscLayoutDestroy(&mat->cmap));
7959: mat->cmap = nmap;
7960: mat->cmap->mapping = l2g;
7961: }
7962: PetscCall(PetscLayoutSetBlockSize(mat->rmap, rbs));
7963: PetscCall(PetscLayoutSetBlockSize(mat->cmap, cbs));
7964: PetscFunctionReturn(PETSC_SUCCESS);
7965: }
7967: /*@
7968: MatSetBlockSizesFromMats - Sets the matrix block row and column sizes to match a pair of matrices
7970: Logically Collective
7972: Input Parameters:
7973: + mat - the matrix
7974: . fromRow - matrix from which to copy row block size
7975: - fromCol - matrix from which to copy column block size (can be same as fromRow)
7977: Level: developer
7979: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`
7980: @*/
7981: PetscErrorCode MatSetBlockSizesFromMats(Mat mat, Mat fromRow, Mat fromCol)
7982: {
7983: PetscFunctionBegin;
7987: if (fromRow->rmap->bs > 0) PetscCall(PetscLayoutSetBlockSize(mat->rmap, fromRow->rmap->bs));
7988: if (fromCol->cmap->bs > 0) PetscCall(PetscLayoutSetBlockSize(mat->cmap, fromCol->cmap->bs));
7989: PetscFunctionReturn(PETSC_SUCCESS);
7990: }
7992: /*@
7993: MatResidual - Default routine to calculate the residual r = b - Ax
7995: Collective
7997: Input Parameters:
7998: + mat - the matrix
7999: . b - the right-hand-side
8000: - x - the approximate solution
8002: Output Parameter:
8003: . r - location to store the residual
8005: Level: developer
8007: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultAdd()`, `PCMGSetResidual()`
8008: @*/
8009: PetscErrorCode MatResidual(Mat mat, Vec b, Vec x, Vec r)
8010: {
8011: PetscFunctionBegin;
8017: MatCheckPreallocated(mat, 1);
8018: PetscCall(PetscLogEventBegin(MAT_Residual, mat, 0, 0, 0));
8019: if (!mat->ops->residual) {
8020: PetscCall(MatMult(mat, x, r));
8021: PetscCall(VecAYPX(r, -1.0, b));
8022: } else {
8023: PetscUseTypeMethod(mat, residual, b, x, r);
8024: }
8025: PetscCall(PetscLogEventEnd(MAT_Residual, mat, 0, 0, 0));
8026: PetscFunctionReturn(PETSC_SUCCESS);
8027: }
8029: /*MC
8030: MatGetRowIJF90 - Obtains the compressed row storage i and j indices for the local rows of a sparse matrix
8032: Synopsis:
8033: MatGetRowIJF90(Mat A, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt n, {PetscInt, pointer :: ia(:)}, {PetscInt, pointer :: ja(:)}, PetscBool done,integer ierr)
8035: Not Collective
8037: Input Parameters:
8038: + A - the matrix
8039: . shift - 0 or 1 indicating we want the indices starting at 0 or 1
8040: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8041: - inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8042: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8043: always used.
8045: Output Parameters:
8046: + n - number of local rows in the (possibly compressed) matrix
8047: . ia - the row pointers; that is ia[0] = 0, ia[row] = ia[row-1] + number of elements in that row of the matrix
8048: . ja - the column indices
8049: - done - indicates if the routine actually worked and returned appropriate ia[] and ja[] arrays; callers
8050: are responsible for handling the case when done == `PETSC_FALSE` and ia and ja are not set
8052: Level: developer
8054: Note:
8055: Use `MatRestoreRowIJF90()` when you no longer need access to the data
8057: .seealso: [](ch_matrices), [](sec_fortranarrays), `Mat`, `MATMPIAIJ`, `MatGetRowIJ()`, `MatRestoreRowIJ()`, `MatRestoreRowIJF90()`
8058: M*/
8060: /*MC
8061: MatRestoreRowIJF90 - restores the compressed row storage i and j indices for the local rows of a sparse matrix obtained with `MatGetRowIJF90()`
8063: Synopsis:
8064: MatRestoreRowIJF90(Mat A, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt n, {PetscInt, pointer :: ia(:)}, {PetscInt, pointer :: ja(:)}, PetscBool done,integer ierr)
8066: Not Collective
8068: Input Parameters:
8069: + A - the matrix
8070: . shift - 0 or 1 indicating we want the indices starting at 0 or 1
8071: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8072: inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8073: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8074: always used.
8075: . n - number of local rows in the (possibly compressed) matrix
8076: . ia - the row pointers; that is ia[0] = 0, ia[row] = ia[row-1] + number of elements in that row of the matrix
8077: . ja - the column indices
8078: - done - indicates if the routine actually worked and returned appropriate ia[] and ja[] arrays; callers
8079: are responsible for handling the case when done == `PETSC_FALSE` and ia and ja are not set
8081: Level: developer
8083: .seealso: [](ch_matrices), [](sec_fortranarrays), `Mat`, `MATMPIAIJ`, `MatGetRowIJ()`, `MatRestoreRowIJ()`, `MatGetRowIJF90()`
8084: M*/
8086: /*@C
8087: MatGetRowIJ - Returns the compressed row storage i and j indices for the local rows of a sparse matrix
8089: Collective
8091: Input Parameters:
8092: + mat - the matrix
8093: . shift - 0 or 1 indicating we want the indices starting at 0 or 1
8094: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8095: - inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8096: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8097: always used.
8099: Output Parameters:
8100: + n - number of local rows in the (possibly compressed) matrix, use `NULL` if not needed
8101: . 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
8102: . ja - the column indices, use `NULL` if not needed
8103: - done - indicates if the routine actually worked and returned appropriate ia[] and ja[] arrays; callers
8104: are responsible for handling the case when done == `PETSC_FALSE` and ia and ja are not set
8106: Level: developer
8108: Notes:
8109: You CANNOT change any of the ia[] or ja[] values.
8111: Use `MatRestoreRowIJ()` when you are finished accessing the ia[] and ja[] values.
8113: Fortran Notes:
8114: Use
8115: .vb
8116: PetscInt, pointer :: ia(:),ja(:)
8117: call MatGetRowIJF90(mat,shift,symmetric,inodecompressed,n,ia,ja,done,ierr)
8118: ! Access the ith and jth entries via ia(i) and ja(j)
8119: .ve
8121: `MatGetRowIJ()` Fortran binding is deprecated (since PETSc 3.19), use `MatGetRowIJF90()`
8123: .seealso: [](ch_matrices), `Mat`, `MATAIJ`, `MatGetRowIJF90()`, `MatGetColumnIJ()`, `MatRestoreRowIJ()`, `MatSeqAIJGetArray()`
8124: @*/
8125: PetscErrorCode MatGetRowIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8126: {
8127: PetscFunctionBegin;
8130: if (n) PetscAssertPointer(n, 5);
8131: if (ia) PetscAssertPointer(ia, 6);
8132: if (ja) PetscAssertPointer(ja, 7);
8133: if (done) PetscAssertPointer(done, 8);
8134: MatCheckPreallocated(mat, 1);
8135: if (!mat->ops->getrowij && done) *done = PETSC_FALSE;
8136: else {
8137: if (done) *done = PETSC_TRUE;
8138: PetscCall(PetscLogEventBegin(MAT_GetRowIJ, mat, 0, 0, 0));
8139: PetscUseTypeMethod(mat, getrowij, shift, symmetric, inodecompressed, n, ia, ja, done);
8140: PetscCall(PetscLogEventEnd(MAT_GetRowIJ, mat, 0, 0, 0));
8141: }
8142: PetscFunctionReturn(PETSC_SUCCESS);
8143: }
8145: /*@C
8146: MatGetColumnIJ - Returns the compressed column storage i and j indices for sequential matrices.
8148: Collective
8150: Input Parameters:
8151: + mat - the matrix
8152: . shift - 1 or zero indicating we want the indices starting at 0 or 1
8153: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be
8154: symmetrized
8155: . inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8156: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8157: always used.
8158: . n - number of columns in the (possibly compressed) matrix
8159: . ia - the column pointers; that is ia[0] = 0, ia[col] = i[col-1] + number of elements in that col of the matrix
8160: - ja - the row indices
8162: Output Parameter:
8163: . done - `PETSC_TRUE` or `PETSC_FALSE`, indicating whether the values have been returned
8165: Level: developer
8167: .seealso: [](ch_matrices), `Mat`, `MatGetRowIJ()`, `MatRestoreColumnIJ()`
8168: @*/
8169: PetscErrorCode MatGetColumnIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8170: {
8171: PetscFunctionBegin;
8174: PetscAssertPointer(n, 5);
8175: if (ia) PetscAssertPointer(ia, 6);
8176: if (ja) PetscAssertPointer(ja, 7);
8177: PetscAssertPointer(done, 8);
8178: MatCheckPreallocated(mat, 1);
8179: if (!mat->ops->getcolumnij) *done = PETSC_FALSE;
8180: else {
8181: *done = PETSC_TRUE;
8182: PetscUseTypeMethod(mat, getcolumnij, shift, symmetric, inodecompressed, n, ia, ja, done);
8183: }
8184: PetscFunctionReturn(PETSC_SUCCESS);
8185: }
8187: /*@C
8188: MatRestoreRowIJ - Call after you are completed with the ia,ja indices obtained with `MatGetRowIJ()`.
8190: Collective
8192: Input Parameters:
8193: + mat - the matrix
8194: . shift - 1 or zero indicating we want the indices starting at 0 or 1
8195: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8196: . inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8197: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8198: always used.
8199: . n - size of (possibly compressed) matrix
8200: . ia - the row pointers
8201: - ja - the column indices
8203: Output Parameter:
8204: . done - `PETSC_TRUE` or `PETSC_FALSE` indicated that the values have been returned
8206: Level: developer
8208: Note:
8209: This routine zeros out `n`, `ia`, and `ja`. This is to prevent accidental
8210: us of the array after it has been restored. If you pass `NULL`, it will
8211: not zero the pointers. Use of ia or ja after `MatRestoreRowIJ()` is invalid.
8213: Fortran Note:
8214: `MatRestoreRowIJ()` Fortran binding is deprecated (since PETSc 3.19), use `MatRestoreRowIJF90()`
8216: .seealso: [](ch_matrices), `Mat`, `MatGetRowIJ()`, `MatRestoreRowIJF90()`, `MatRestoreColumnIJ()`
8217: @*/
8218: PetscErrorCode MatRestoreRowIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8219: {
8220: PetscFunctionBegin;
8223: if (ia) PetscAssertPointer(ia, 6);
8224: if (ja) PetscAssertPointer(ja, 7);
8225: if (done) PetscAssertPointer(done, 8);
8226: MatCheckPreallocated(mat, 1);
8228: if (!mat->ops->restorerowij && done) *done = PETSC_FALSE;
8229: else {
8230: if (done) *done = PETSC_TRUE;
8231: PetscUseTypeMethod(mat, restorerowij, shift, symmetric, inodecompressed, n, ia, ja, done);
8232: if (n) *n = 0;
8233: if (ia) *ia = NULL;
8234: if (ja) *ja = NULL;
8235: }
8236: PetscFunctionReturn(PETSC_SUCCESS);
8237: }
8239: /*@C
8240: MatRestoreColumnIJ - Call after you are completed with the ia,ja indices obtained with `MatGetColumnIJ()`.
8242: Collective
8244: Input Parameters:
8245: + mat - the matrix
8246: . shift - 1 or zero indicating we want the indices starting at 0 or 1
8247: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8248: - inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8249: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8250: always used.
8252: Output Parameters:
8253: + n - size of (possibly compressed) matrix
8254: . ia - the column pointers
8255: . ja - the row indices
8256: - done - `PETSC_TRUE` or `PETSC_FALSE` indicated that the values have been returned
8258: Level: developer
8260: .seealso: [](ch_matrices), `Mat`, `MatGetColumnIJ()`, `MatRestoreRowIJ()`
8261: @*/
8262: PetscErrorCode MatRestoreColumnIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8263: {
8264: PetscFunctionBegin;
8267: if (ia) PetscAssertPointer(ia, 6);
8268: if (ja) PetscAssertPointer(ja, 7);
8269: PetscAssertPointer(done, 8);
8270: MatCheckPreallocated(mat, 1);
8272: if (!mat->ops->restorecolumnij) *done = PETSC_FALSE;
8273: else {
8274: *done = PETSC_TRUE;
8275: PetscUseTypeMethod(mat, restorecolumnij, shift, symmetric, inodecompressed, n, ia, ja, done);
8276: if (n) *n = 0;
8277: if (ia) *ia = NULL;
8278: if (ja) *ja = NULL;
8279: }
8280: PetscFunctionReturn(PETSC_SUCCESS);
8281: }
8283: /*@
8284: MatColoringPatch - Used inside matrix coloring routines that use `MatGetRowIJ()` and/or
8285: `MatGetColumnIJ()`.
8287: Collective
8289: Input Parameters:
8290: + mat - the matrix
8291: . ncolors - maximum color value
8292: . n - number of entries in colorarray
8293: - colorarray - array indicating color for each column
8295: Output Parameter:
8296: . iscoloring - coloring generated using colorarray information
8298: Level: developer
8300: .seealso: [](ch_matrices), `Mat`, `MatGetRowIJ()`, `MatGetColumnIJ()`
8301: @*/
8302: PetscErrorCode MatColoringPatch(Mat mat, PetscInt ncolors, PetscInt n, ISColoringValue colorarray[], ISColoring *iscoloring)
8303: {
8304: PetscFunctionBegin;
8307: PetscAssertPointer(colorarray, 4);
8308: PetscAssertPointer(iscoloring, 5);
8309: MatCheckPreallocated(mat, 1);
8311: if (!mat->ops->coloringpatch) {
8312: PetscCall(ISColoringCreate(PetscObjectComm((PetscObject)mat), ncolors, n, colorarray, PETSC_OWN_POINTER, iscoloring));
8313: } else {
8314: PetscUseTypeMethod(mat, coloringpatch, ncolors, n, colorarray, iscoloring);
8315: }
8316: PetscFunctionReturn(PETSC_SUCCESS);
8317: }
8319: /*@
8320: MatSetUnfactored - Resets a factored matrix to be treated as unfactored.
8322: Logically Collective
8324: Input Parameter:
8325: . mat - the factored matrix to be reset
8327: Level: developer
8329: Notes:
8330: This routine should be used only with factored matrices formed by in-place
8331: factorization via ILU(0) (or by in-place LU factorization for the `MATSEQDENSE`
8332: format). This option can save memory, for example, when solving nonlinear
8333: systems with a matrix-free Newton-Krylov method and a matrix-based, in-place
8334: ILU(0) preconditioner.
8336: One can specify in-place ILU(0) factorization by calling
8337: .vb
8338: PCType(pc,PCILU);
8339: PCFactorSeUseInPlace(pc);
8340: .ve
8341: or by using the options -pc_type ilu -pc_factor_in_place
8343: In-place factorization ILU(0) can also be used as a local
8344: solver for the blocks within the block Jacobi or additive Schwarz
8345: methods (runtime option: -sub_pc_factor_in_place). See Users-Manual: ch_pc
8346: for details on setting local solver options.
8348: Most users should employ the `KSP` interface for linear solvers
8349: instead of working directly with matrix algebra routines such as this.
8350: See, e.g., `KSPCreate()`.
8352: .seealso: [](ch_matrices), `Mat`, `PCFactorSetUseInPlace()`, `PCFactorGetUseInPlace()`
8353: @*/
8354: PetscErrorCode MatSetUnfactored(Mat mat)
8355: {
8356: PetscFunctionBegin;
8359: MatCheckPreallocated(mat, 1);
8360: mat->factortype = MAT_FACTOR_NONE;
8361: if (!mat->ops->setunfactored) PetscFunctionReturn(PETSC_SUCCESS);
8362: PetscUseTypeMethod(mat, setunfactored);
8363: PetscFunctionReturn(PETSC_SUCCESS);
8364: }
8366: /*MC
8367: MatDenseGetArrayF90 - Accesses a matrix array from Fortran
8369: Synopsis:
8370: MatDenseGetArrayF90(Mat x,{Scalar, pointer :: xx_v(:,:)},integer ierr)
8372: Not Collective
8374: Input Parameter:
8375: . x - matrix
8377: Output Parameters:
8378: + xx_v - the Fortran pointer to the array
8379: - ierr - error code
8381: Example of Usage:
8382: .vb
8383: PetscScalar, pointer xx_v(:,:)
8384: ....
8385: call MatDenseGetArrayF90(x,xx_v,ierr)
8386: a = xx_v(3)
8387: call MatDenseRestoreArrayF90(x,xx_v,ierr)
8388: .ve
8390: Level: advanced
8392: .seealso: [](ch_matrices), `Mat`, `MatDenseRestoreArrayF90()`, `MatDenseGetArray()`, `MatDenseRestoreArray()`, `MatSeqAIJGetArrayF90()`
8393: M*/
8395: /*MC
8396: MatDenseRestoreArrayF90 - Restores a matrix array that has been
8397: accessed with `MatDenseGetArrayF90()`.
8399: Synopsis:
8400: MatDenseRestoreArrayF90(Mat x,{Scalar, pointer :: xx_v(:,:)},integer ierr)
8402: Not Collective
8404: Input Parameters:
8405: + x - matrix
8406: - xx_v - the Fortran90 pointer to the array
8408: Output Parameter:
8409: . ierr - error code
8411: Example of Usage:
8412: .vb
8413: PetscScalar, pointer xx_v(:,:)
8414: ....
8415: call MatDenseGetArrayF90(x,xx_v,ierr)
8416: a = xx_v(3)
8417: call MatDenseRestoreArrayF90(x,xx_v,ierr)
8418: .ve
8420: Level: advanced
8422: .seealso: [](ch_matrices), `Mat`, `MatDenseGetArrayF90()`, `MatDenseGetArray()`, `MatDenseRestoreArray()`, `MatSeqAIJRestoreArrayF90()`
8423: M*/
8425: /*MC
8426: MatSeqAIJGetArrayF90 - Accesses a matrix array from Fortran.
8428: Synopsis:
8429: MatSeqAIJGetArrayF90(Mat x,{Scalar, pointer :: xx_v(:)},integer ierr)
8431: Not Collective
8433: Input Parameter:
8434: . x - matrix
8436: Output Parameters:
8437: + xx_v - the Fortran pointer to the array
8438: - ierr - error code
8440: Example of Usage:
8441: .vb
8442: PetscScalar, pointer xx_v(:)
8443: ....
8444: call MatSeqAIJGetArrayF90(x,xx_v,ierr)
8445: a = xx_v(3)
8446: call MatSeqAIJRestoreArrayF90(x,xx_v,ierr)
8447: .ve
8449: Level: advanced
8451: .seealso: [](ch_matrices), `Mat`, `MatSeqAIJRestoreArrayF90()`, `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArray()`, `MatDenseGetArrayF90()`
8452: M*/
8454: /*MC
8455: MatSeqAIJRestoreArrayF90 - Restores a matrix array that has been
8456: accessed with `MatSeqAIJGetArrayF90()`.
8458: Synopsis:
8459: MatSeqAIJRestoreArrayF90(Mat x,{Scalar, pointer :: xx_v(:)},integer ierr)
8461: Not Collective
8463: Input Parameters:
8464: + x - matrix
8465: - xx_v - the Fortran90 pointer to the array
8467: Output Parameter:
8468: . ierr - error code
8470: Example of Usage:
8471: .vb
8472: PetscScalar, pointer xx_v(:)
8473: ....
8474: call MatSeqAIJGetArrayF90(x,xx_v,ierr)
8475: a = xx_v(3)
8476: call MatSeqAIJRestoreArrayF90(x,xx_v,ierr)
8477: .ve
8479: Level: advanced
8481: .seealso: [](ch_matrices), `Mat`, `MatSeqAIJGetArrayF90()`, `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArray()`, `MatDenseRestoreArrayF90()`
8482: M*/
8484: /*@
8485: MatCreateSubMatrix - Gets a single submatrix on the same number of processors
8486: as the original matrix.
8488: Collective
8490: Input Parameters:
8491: + mat - the original matrix
8492: . isrow - parallel `IS` containing the rows this processor should obtain
8493: . 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.
8494: - cll - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
8496: Output Parameter:
8497: . newmat - the new submatrix, of the same type as the original matrix
8499: Level: advanced
8501: Notes:
8502: The submatrix will be able to be multiplied with vectors using the same layout as `iscol`.
8504: Some matrix types place restrictions on the row and column indices, such
8505: 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;
8506: for example, if the block size is 3 one cannot select the 0 and 2 rows without selecting the 1 row.
8508: The index sets may not have duplicate entries.
8510: The first time this is called you should use a cll of `MAT_INITIAL_MATRIX`,
8511: the `MatCreateSubMatrix()` routine will create the newmat for you. Any additional calls
8512: to this routine with a mat of the same nonzero structure and with a call of `MAT_REUSE_MATRIX`
8513: will reuse the matrix generated the first time. You should call `MatDestroy()` on `newmat` when
8514: you are finished using it.
8516: The communicator of the newly obtained matrix is ALWAYS the same as the communicator of
8517: the input matrix.
8519: If `iscol` is `NULL` then all columns are obtained (not supported in Fortran).
8521: If `isrow` and `iscol` have a nontrivial block-size, then the resulting matrix has this block-size as well. This feature
8522: is used by `PCFIELDSPLIT` to allow easy nesting of its use.
8524: Example usage:
8525: Consider the following 8x8 matrix with 34 non-zero values, that is
8526: assembled across 3 processors. Let's assume that proc0 owns 3 rows,
8527: proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown
8528: as follows
8529: .vb
8530: 1 2 0 | 0 3 0 | 0 4
8531: Proc0 0 5 6 | 7 0 0 | 8 0
8532: 9 0 10 | 11 0 0 | 12 0
8533: -------------------------------------
8534: 13 0 14 | 15 16 17 | 0 0
8535: Proc1 0 18 0 | 19 20 21 | 0 0
8536: 0 0 0 | 22 23 0 | 24 0
8537: -------------------------------------
8538: Proc2 25 26 27 | 0 0 28 | 29 0
8539: 30 0 0 | 31 32 33 | 0 34
8540: .ve
8542: Suppose `isrow` = [0 1 | 4 | 6 7] and `iscol` = [1 2 | 3 4 5 | 6]. The resulting submatrix is
8544: .vb
8545: 2 0 | 0 3 0 | 0
8546: Proc0 5 6 | 7 0 0 | 8
8547: -------------------------------
8548: Proc1 18 0 | 19 20 21 | 0
8549: -------------------------------
8550: Proc2 26 27 | 0 0 28 | 29
8551: 0 0 | 31 32 33 | 0
8552: .ve
8554: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatCreateSubMatricesMPI()`, `MatCreateSubMatrixVirtual()`, `MatSubMatrixVirtualUpdate()`
8555: @*/
8556: PetscErrorCode MatCreateSubMatrix(Mat mat, IS isrow, IS iscol, MatReuse cll, Mat *newmat)
8557: {
8558: PetscMPIInt size;
8559: Mat *local;
8560: IS iscoltmp;
8561: PetscBool flg;
8563: PetscFunctionBegin;
8567: PetscAssertPointer(newmat, 5);
8570: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
8571: PetscCheck(cll != MAT_IGNORE_MATRIX, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot use MAT_IGNORE_MATRIX");
8573: MatCheckPreallocated(mat, 1);
8574: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
8576: if (!iscol || isrow == iscol) {
8577: PetscBool stride;
8578: PetscMPIInt grabentirematrix = 0, grab;
8579: PetscCall(PetscObjectTypeCompare((PetscObject)isrow, ISSTRIDE, &stride));
8580: if (stride) {
8581: PetscInt first, step, n, rstart, rend;
8582: PetscCall(ISStrideGetInfo(isrow, &first, &step));
8583: if (step == 1) {
8584: PetscCall(MatGetOwnershipRange(mat, &rstart, &rend));
8585: if (rstart == first) {
8586: PetscCall(ISGetLocalSize(isrow, &n));
8587: if (n == rend - rstart) grabentirematrix = 1;
8588: }
8589: }
8590: }
8591: PetscCallMPI(MPIU_Allreduce(&grabentirematrix, &grab, 1, MPI_INT, MPI_MIN, PetscObjectComm((PetscObject)mat)));
8592: if (grab) {
8593: PetscCall(PetscInfo(mat, "Getting entire matrix as submatrix\n"));
8594: if (cll == MAT_INITIAL_MATRIX) {
8595: *newmat = mat;
8596: PetscCall(PetscObjectReference((PetscObject)mat));
8597: }
8598: PetscFunctionReturn(PETSC_SUCCESS);
8599: }
8600: }
8602: if (!iscol) {
8603: PetscCall(ISCreateStride(PetscObjectComm((PetscObject)mat), mat->cmap->n, mat->cmap->rstart, 1, &iscoltmp));
8604: } else {
8605: iscoltmp = iscol;
8606: }
8608: /* if original matrix is on just one processor then use submatrix generated */
8609: if (mat->ops->createsubmatrices && !mat->ops->createsubmatrix && size == 1 && cll == MAT_REUSE_MATRIX) {
8610: PetscCall(MatCreateSubMatrices(mat, 1, &isrow, &iscoltmp, MAT_REUSE_MATRIX, &newmat));
8611: goto setproperties;
8612: } else if (mat->ops->createsubmatrices && !mat->ops->createsubmatrix && size == 1) {
8613: PetscCall(MatCreateSubMatrices(mat, 1, &isrow, &iscoltmp, MAT_INITIAL_MATRIX, &local));
8614: *newmat = *local;
8615: PetscCall(PetscFree(local));
8616: goto setproperties;
8617: } else if (!mat->ops->createsubmatrix) {
8618: /* Create a new matrix type that implements the operation using the full matrix */
8619: PetscCall(PetscLogEventBegin(MAT_CreateSubMat, mat, 0, 0, 0));
8620: switch (cll) {
8621: case MAT_INITIAL_MATRIX:
8622: PetscCall(MatCreateSubMatrixVirtual(mat, isrow, iscoltmp, newmat));
8623: break;
8624: case MAT_REUSE_MATRIX:
8625: PetscCall(MatSubMatrixVirtualUpdate(*newmat, mat, isrow, iscoltmp));
8626: break;
8627: default:
8628: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Invalid MatReuse, must be either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX");
8629: }
8630: PetscCall(PetscLogEventEnd(MAT_CreateSubMat, mat, 0, 0, 0));
8631: goto setproperties;
8632: }
8634: PetscCall(PetscLogEventBegin(MAT_CreateSubMat, mat, 0, 0, 0));
8635: PetscUseTypeMethod(mat, createsubmatrix, isrow, iscoltmp, cll, newmat);
8636: PetscCall(PetscLogEventEnd(MAT_CreateSubMat, mat, 0, 0, 0));
8638: setproperties:
8639: if ((*newmat)->symmetric == PETSC_BOOL3_UNKNOWN && (*newmat)->structurally_symmetric == PETSC_BOOL3_UNKNOWN && (*newmat)->spd == PETSC_BOOL3_UNKNOWN && (*newmat)->hermitian == PETSC_BOOL3_UNKNOWN) {
8640: PetscCall(ISEqualUnsorted(isrow, iscoltmp, &flg));
8641: if (flg) PetscCall(MatPropagateSymmetryOptions(mat, *newmat));
8642: }
8643: if (!iscol) PetscCall(ISDestroy(&iscoltmp));
8644: if (*newmat && cll == MAT_INITIAL_MATRIX) PetscCall(PetscObjectStateIncrease((PetscObject)*newmat));
8645: PetscFunctionReturn(PETSC_SUCCESS);
8646: }
8648: /*@
8649: MatPropagateSymmetryOptions - Propagates symmetry options set on a matrix to another matrix
8651: Not Collective
8653: Input Parameters:
8654: + A - the matrix we wish to propagate options from
8655: - B - the matrix we wish to propagate options to
8657: Level: beginner
8659: Note:
8660: Propagates the options associated to `MAT_SYMMETRY_ETERNAL`, `MAT_STRUCTURALLY_SYMMETRIC`, `MAT_HERMITIAN`, `MAT_SPD`, `MAT_SYMMETRIC`, and `MAT_STRUCTURAL_SYMMETRY_ETERNAL`
8662: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatIsSymmetricKnown()`, `MatIsSPDKnown()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetricKnown()`
8663: @*/
8664: PetscErrorCode MatPropagateSymmetryOptions(Mat A, Mat B)
8665: {
8666: PetscFunctionBegin;
8669: B->symmetry_eternal = A->symmetry_eternal;
8670: B->structural_symmetry_eternal = A->structural_symmetry_eternal;
8671: B->symmetric = A->symmetric;
8672: B->structurally_symmetric = A->structurally_symmetric;
8673: B->spd = A->spd;
8674: B->hermitian = A->hermitian;
8675: PetscFunctionReturn(PETSC_SUCCESS);
8676: }
8678: /*@
8679: MatStashSetInitialSize - sets the sizes of the matrix stash, that is
8680: used during the assembly process to store values that belong to
8681: other processors.
8683: Not Collective
8685: Input Parameters:
8686: + mat - the matrix
8687: . size - the initial size of the stash.
8688: - bsize - the initial size of the block-stash(if used).
8690: Options Database Keys:
8691: + -matstash_initial_size <size> or <size0,size1,...sizep-1> - set initial size
8692: - -matstash_block_initial_size <bsize> or <bsize0,bsize1,...bsizep-1> - set initial block size
8694: Level: intermediate
8696: Notes:
8697: The block-stash is used for values set with `MatSetValuesBlocked()` while
8698: the stash is used for values set with `MatSetValues()`
8700: Run with the option -info and look for output of the form
8701: MatAssemblyBegin_MPIXXX:Stash has MM entries, uses nn mallocs.
8702: to determine the appropriate value, MM, to use for size and
8703: MatAssemblyBegin_MPIXXX:Block-Stash has BMM entries, uses nn mallocs.
8704: to determine the value, BMM to use for bsize
8706: .seealso: [](ch_matrices), `MatAssemblyBegin()`, `MatAssemblyEnd()`, `Mat`, `MatStashGetInfo()`
8707: @*/
8708: PetscErrorCode MatStashSetInitialSize(Mat mat, PetscInt size, PetscInt bsize)
8709: {
8710: PetscFunctionBegin;
8713: PetscCall(MatStashSetInitialSize_Private(&mat->stash, size));
8714: PetscCall(MatStashSetInitialSize_Private(&mat->bstash, bsize));
8715: PetscFunctionReturn(PETSC_SUCCESS);
8716: }
8718: /*@
8719: MatInterpolateAdd - $w = y + A*x$ or $A^T*x$ depending on the shape of
8720: the matrix
8722: Neighbor-wise Collective
8724: Input Parameters:
8725: + A - the matrix
8726: . x - the vector to be multiplied by the interpolation operator
8727: - y - the vector to be added to the result
8729: Output Parameter:
8730: . w - the resulting vector
8732: Level: intermediate
8734: Notes:
8735: `w` may be the same vector as `y`.
8737: This allows one to use either the restriction or interpolation (its transpose)
8738: matrix to do the interpolation
8740: .seealso: [](ch_matrices), `Mat`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatRestrict()`, `PCMG`
8741: @*/
8742: PetscErrorCode MatInterpolateAdd(Mat A, Vec x, Vec y, Vec w)
8743: {
8744: PetscInt M, N, Ny;
8746: PetscFunctionBegin;
8751: PetscCall(MatGetSize(A, &M, &N));
8752: PetscCall(VecGetSize(y, &Ny));
8753: if (M == Ny) {
8754: PetscCall(MatMultAdd(A, x, y, w));
8755: } else {
8756: PetscCall(MatMultTransposeAdd(A, x, y, w));
8757: }
8758: PetscFunctionReturn(PETSC_SUCCESS);
8759: }
8761: /*@
8762: MatInterpolate - $y = A*x$ or $A^T*x$ depending on the shape of
8763: the matrix
8765: Neighbor-wise Collective
8767: Input Parameters:
8768: + A - the matrix
8769: - x - the vector to be interpolated
8771: Output Parameter:
8772: . y - the resulting vector
8774: Level: intermediate
8776: Note:
8777: This allows one to use either the restriction or interpolation (its transpose)
8778: matrix to do the interpolation
8780: .seealso: [](ch_matrices), `Mat`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatRestrict()`, `PCMG`
8781: @*/
8782: PetscErrorCode MatInterpolate(Mat A, Vec x, Vec y)
8783: {
8784: PetscInt M, N, Ny;
8786: PetscFunctionBegin;
8790: PetscCall(MatGetSize(A, &M, &N));
8791: PetscCall(VecGetSize(y, &Ny));
8792: if (M == Ny) {
8793: PetscCall(MatMult(A, x, y));
8794: } else {
8795: PetscCall(MatMultTranspose(A, x, y));
8796: }
8797: PetscFunctionReturn(PETSC_SUCCESS);
8798: }
8800: /*@
8801: MatRestrict - $y = A*x$ or $A^T*x$
8803: Neighbor-wise Collective
8805: Input Parameters:
8806: + A - the matrix
8807: - x - the vector to be restricted
8809: Output Parameter:
8810: . y - the resulting vector
8812: Level: intermediate
8814: Note:
8815: This allows one to use either the restriction or interpolation (its transpose)
8816: matrix to do the restriction
8818: .seealso: [](ch_matrices), `Mat`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatInterpolate()`, `PCMG`
8819: @*/
8820: PetscErrorCode MatRestrict(Mat A, Vec x, Vec y)
8821: {
8822: PetscInt M, N, Nx;
8824: PetscFunctionBegin;
8828: PetscCall(MatGetSize(A, &M, &N));
8829: PetscCall(VecGetSize(x, &Nx));
8830: if (M == Nx) {
8831: PetscCall(MatMultTranspose(A, x, y));
8832: } else {
8833: PetscCall(MatMult(A, x, y));
8834: }
8835: PetscFunctionReturn(PETSC_SUCCESS);
8836: }
8838: /*@
8839: MatMatInterpolateAdd - $Y = W + A*X$ or $W + A^T*X$ depending on the shape of `A`
8841: Neighbor-wise Collective
8843: Input Parameters:
8844: + A - the matrix
8845: . x - the input dense matrix to be multiplied
8846: - w - the input dense matrix to be added to the result
8848: Output Parameter:
8849: . y - the output dense matrix
8851: Level: intermediate
8853: Note:
8854: This allows one to use either the restriction or interpolation (its transpose)
8855: matrix to do the interpolation. `y` matrix can be reused if already created with the proper sizes,
8856: otherwise it will be recreated. `y` must be initialized to `NULL` if not supplied.
8858: .seealso: [](ch_matrices), `Mat`, `MatInterpolateAdd()`, `MatMatInterpolate()`, `MatMatRestrict()`, `PCMG`
8859: @*/
8860: PetscErrorCode MatMatInterpolateAdd(Mat A, Mat x, Mat w, Mat *y)
8861: {
8862: PetscInt M, N, Mx, Nx, Mo, My = 0, Ny = 0;
8863: PetscBool trans = PETSC_TRUE;
8864: MatReuse reuse = MAT_INITIAL_MATRIX;
8866: PetscFunctionBegin;
8872: PetscCall(MatGetSize(A, &M, &N));
8873: PetscCall(MatGetSize(x, &Mx, &Nx));
8874: if (N == Mx) trans = PETSC_FALSE;
8875: 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);
8876: Mo = trans ? N : M;
8877: if (*y) {
8878: PetscCall(MatGetSize(*y, &My, &Ny));
8879: if (Mo == My && Nx == Ny) {
8880: reuse = MAT_REUSE_MATRIX;
8881: } else {
8882: 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);
8883: PetscCall(MatDestroy(y));
8884: }
8885: }
8887: if (w && *y == w) { /* this is to minimize changes in PCMG */
8888: PetscBool flg;
8890: PetscCall(PetscObjectQuery((PetscObject)*y, "__MatMatIntAdd_w", (PetscObject *)&w));
8891: if (w) {
8892: PetscInt My, Ny, Mw, Nw;
8894: PetscCall(PetscObjectTypeCompare((PetscObject)*y, ((PetscObject)w)->type_name, &flg));
8895: PetscCall(MatGetSize(*y, &My, &Ny));
8896: PetscCall(MatGetSize(w, &Mw, &Nw));
8897: if (!flg || My != Mw || Ny != Nw) w = NULL;
8898: }
8899: if (!w) {
8900: PetscCall(MatDuplicate(*y, MAT_COPY_VALUES, &w));
8901: PetscCall(PetscObjectCompose((PetscObject)*y, "__MatMatIntAdd_w", (PetscObject)w));
8902: PetscCall(PetscObjectDereference((PetscObject)w));
8903: } else {
8904: PetscCall(MatCopy(*y, w, UNKNOWN_NONZERO_PATTERN));
8905: }
8906: }
8907: if (!trans) {
8908: PetscCall(MatMatMult(A, x, reuse, PETSC_DETERMINE, y));
8909: } else {
8910: PetscCall(MatTransposeMatMult(A, x, reuse, PETSC_DETERMINE, y));
8911: }
8912: if (w) PetscCall(MatAXPY(*y, 1.0, w, UNKNOWN_NONZERO_PATTERN));
8913: PetscFunctionReturn(PETSC_SUCCESS);
8914: }
8916: /*@
8917: MatMatInterpolate - $Y = A*X$ or $A^T*X$ depending on the shape of `A`
8919: Neighbor-wise Collective
8921: Input Parameters:
8922: + A - the matrix
8923: - x - the input dense matrix
8925: Output Parameter:
8926: . y - the output dense matrix
8928: Level: intermediate
8930: Note:
8931: This allows one to use either the restriction or interpolation (its transpose)
8932: matrix to do the interpolation. `y` matrix can be reused if already created with the proper sizes,
8933: otherwise it will be recreated. `y` must be initialized to `NULL` if not supplied.
8935: .seealso: [](ch_matrices), `Mat`, `MatInterpolate()`, `MatRestrict()`, `MatMatRestrict()`, `PCMG`
8936: @*/
8937: PetscErrorCode MatMatInterpolate(Mat A, Mat x, Mat *y)
8938: {
8939: PetscFunctionBegin;
8940: PetscCall(MatMatInterpolateAdd(A, x, NULL, y));
8941: PetscFunctionReturn(PETSC_SUCCESS);
8942: }
8944: /*@
8945: MatMatRestrict - $Y = A*X$ or $A^T*X$ depending on the shape of `A`
8947: Neighbor-wise Collective
8949: Input Parameters:
8950: + A - the matrix
8951: - x - the input dense matrix
8953: Output Parameter:
8954: . y - the output dense matrix
8956: Level: intermediate
8958: Note:
8959: This allows one to use either the restriction or interpolation (its transpose)
8960: matrix to do the restriction. `y` matrix can be reused if already created with the proper sizes,
8961: otherwise it will be recreated. `y` must be initialized to `NULL` if not supplied.
8963: .seealso: [](ch_matrices), `Mat`, `MatRestrict()`, `MatInterpolate()`, `MatMatInterpolate()`, `PCMG`
8964: @*/
8965: PetscErrorCode MatMatRestrict(Mat A, Mat x, Mat *y)
8966: {
8967: PetscFunctionBegin;
8968: PetscCall(MatMatInterpolateAdd(A, x, NULL, y));
8969: PetscFunctionReturn(PETSC_SUCCESS);
8970: }
8972: /*@
8973: MatGetNullSpace - retrieves the null space of a matrix.
8975: Logically Collective
8977: Input Parameters:
8978: + mat - the matrix
8979: - nullsp - the null space object
8981: Level: developer
8983: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatSetNullSpace()`, `MatNullSpace`
8984: @*/
8985: PetscErrorCode MatGetNullSpace(Mat mat, MatNullSpace *nullsp)
8986: {
8987: PetscFunctionBegin;
8989: PetscAssertPointer(nullsp, 2);
8990: *nullsp = (mat->symmetric == PETSC_BOOL3_TRUE && !mat->nullsp) ? mat->transnullsp : mat->nullsp;
8991: PetscFunctionReturn(PETSC_SUCCESS);
8992: }
8994: /*@C
8995: MatGetNullSpaces - gets the null spaces, transpose null spaces, and near null spaces from an array of matrices
8997: Logically Collective
8999: Input Parameters:
9000: + n - the number of matrices
9001: - mat - the array of matrices
9003: Output Parameters:
9004: . nullsp - an array of null spaces, `NULL` for each matrix that does not have a null space, length 3 * `n`
9006: Level: developer
9008: Note:
9009: Call `MatRestoreNullspaces()` to provide these to another array of matrices
9011: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetTransposeNullSpace()`, `MatGetTransposeNullSpace()`,
9012: `MatNullSpaceRemove()`, `MatRestoreNullSpaces()`
9013: @*/
9014: PetscErrorCode MatGetNullSpaces(PetscInt n, Mat mat[], MatNullSpace *nullsp[])
9015: {
9016: PetscFunctionBegin;
9017: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Number of matrices %" PetscInt_FMT " must be non-negative", n);
9018: PetscAssertPointer(mat, 2);
9019: PetscAssertPointer(nullsp, 3);
9021: PetscCall(PetscCalloc1(3 * n, nullsp));
9022: for (PetscInt i = 0; i < n; i++) {
9024: (*nullsp)[i] = mat[i]->nullsp;
9025: PetscCall(PetscObjectReference((PetscObject)(*nullsp)[i]));
9026: (*nullsp)[n + i] = mat[i]->nearnullsp;
9027: PetscCall(PetscObjectReference((PetscObject)(*nullsp)[n + i]));
9028: (*nullsp)[2 * n + i] = mat[i]->transnullsp;
9029: PetscCall(PetscObjectReference((PetscObject)(*nullsp)[2 * n + i]));
9030: }
9031: PetscFunctionReturn(PETSC_SUCCESS);
9032: }
9034: /*@C
9035: MatRestoreNullSpaces - sets the null spaces, transpose null spaces, and near null spaces obtained with `MatGetNullSpaces()` for an array of matrices
9037: Logically Collective
9039: Input Parameters:
9040: + n - the number of matrices
9041: . mat - the array of matrices
9042: - nullsp - an array of null spaces
9044: Level: developer
9046: Note:
9047: Call `MatGetNullSpaces()` to create `nullsp`
9049: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetTransposeNullSpace()`, `MatGetTransposeNullSpace()`,
9050: `MatNullSpaceRemove()`, `MatGetNullSpaces()`
9051: @*/
9052: PetscErrorCode MatRestoreNullSpaces(PetscInt n, Mat mat[], MatNullSpace *nullsp[])
9053: {
9054: PetscFunctionBegin;
9055: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Number of matrices %" PetscInt_FMT " must be non-negative", n);
9056: PetscAssertPointer(mat, 2);
9057: PetscAssertPointer(nullsp, 3);
9058: PetscAssertPointer(*nullsp, 3);
9060: for (PetscInt i = 0; i < n; i++) {
9062: PetscCall(MatSetNullSpace(mat[i], (*nullsp)[i]));
9063: PetscCall(PetscObjectDereference((PetscObject)(*nullsp)[i]));
9064: PetscCall(MatSetNearNullSpace(mat[i], (*nullsp)[n + i]));
9065: PetscCall(PetscObjectDereference((PetscObject)(*nullsp)[n + i]));
9066: PetscCall(MatSetTransposeNullSpace(mat[i], (*nullsp)[2 * n + i]));
9067: PetscCall(PetscObjectDereference((PetscObject)(*nullsp)[2 * n + i]));
9068: }
9069: PetscCall(PetscFree(*nullsp));
9070: PetscFunctionReturn(PETSC_SUCCESS);
9071: }
9073: /*@
9074: MatSetNullSpace - attaches a null space to a matrix.
9076: Logically Collective
9078: Input Parameters:
9079: + mat - the matrix
9080: - nullsp - the null space object
9082: Level: advanced
9084: Notes:
9085: This null space is used by the `KSP` linear solvers to solve singular systems.
9087: 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`
9089: 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 to
9090: to zero but the linear system will still be solved in a least squares sense.
9092: The fundamental theorem of linear algebra (Gilbert Strang, Introduction to Applied Mathematics, page 72) states that
9093: 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), + the range of $A^T$, $R(A^T)$.
9094: 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
9095: 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
9096: 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($A^T$).
9097: This \hat{b} can be obtained by calling `MatNullSpaceRemove()` with the null space of the transpose of the matrix.
9099: If the matrix is known to be symmetric because it is an `MATSBAIJ` matrix or one as called
9100: `MatSetOption`(mat,`MAT_SYMMETRIC` or possibly `MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`); this
9101: routine also automatically calls `MatSetTransposeNullSpace()`.
9103: The user should call `MatNullSpaceDestroy()`.
9105: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetTransposeNullSpace()`, `MatGetTransposeNullSpace()`, `MatNullSpaceRemove()`,
9106: `KSPSetPCSide()`
9107: @*/
9108: PetscErrorCode MatSetNullSpace(Mat mat, MatNullSpace nullsp)
9109: {
9110: PetscFunctionBegin;
9113: if (nullsp) PetscCall(PetscObjectReference((PetscObject)nullsp));
9114: PetscCall(MatNullSpaceDestroy(&mat->nullsp));
9115: mat->nullsp = nullsp;
9116: if (mat->symmetric == PETSC_BOOL3_TRUE) PetscCall(MatSetTransposeNullSpace(mat, nullsp));
9117: PetscFunctionReturn(PETSC_SUCCESS);
9118: }
9120: /*@
9121: MatGetTransposeNullSpace - retrieves the null space of the transpose of a matrix.
9123: Logically Collective
9125: Input Parameters:
9126: + mat - the matrix
9127: - nullsp - the null space object
9129: Level: developer
9131: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatSetTransposeNullSpace()`, `MatSetNullSpace()`, `MatGetNullSpace()`
9132: @*/
9133: PetscErrorCode MatGetTransposeNullSpace(Mat mat, MatNullSpace *nullsp)
9134: {
9135: PetscFunctionBegin;
9138: PetscAssertPointer(nullsp, 2);
9139: *nullsp = (mat->symmetric == PETSC_BOOL3_TRUE && !mat->transnullsp) ? mat->nullsp : mat->transnullsp;
9140: PetscFunctionReturn(PETSC_SUCCESS);
9141: }
9143: /*@
9144: MatSetTransposeNullSpace - attaches the null space of a transpose of a matrix to the matrix
9146: Logically Collective
9148: Input Parameters:
9149: + mat - the matrix
9150: - nullsp - the null space object
9152: Level: advanced
9154: Notes:
9155: This allows solving singular linear systems defined by the transpose of the matrix using `KSP` solvers with left preconditioning.
9157: See `MatSetNullSpace()`
9159: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetNullSpace()`, `MatGetTransposeNullSpace()`, `MatNullSpaceRemove()`, `KSPSetPCSide()`
9160: @*/
9161: PetscErrorCode MatSetTransposeNullSpace(Mat mat, MatNullSpace nullsp)
9162: {
9163: PetscFunctionBegin;
9166: if (nullsp) PetscCall(PetscObjectReference((PetscObject)nullsp));
9167: PetscCall(MatNullSpaceDestroy(&mat->transnullsp));
9168: mat->transnullsp = nullsp;
9169: PetscFunctionReturn(PETSC_SUCCESS);
9170: }
9172: /*@
9173: MatSetNearNullSpace - attaches a null space to a matrix, which is often the null space (rigid body modes) of the operator without boundary conditions
9174: This null space will be used to provide near null space vectors to a multigrid preconditioner built from this matrix.
9176: Logically Collective
9178: Input Parameters:
9179: + mat - the matrix
9180: - nullsp - the null space object
9182: Level: advanced
9184: Notes:
9185: Overwrites any previous near null space that may have been attached
9187: You can remove the null space by calling this routine with an `nullsp` of `NULL`
9189: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNullSpace()`, `MatNullSpaceCreateRigidBody()`, `MatGetNearNullSpace()`
9190: @*/
9191: PetscErrorCode MatSetNearNullSpace(Mat mat, MatNullSpace nullsp)
9192: {
9193: PetscFunctionBegin;
9197: MatCheckPreallocated(mat, 1);
9198: if (nullsp) PetscCall(PetscObjectReference((PetscObject)nullsp));
9199: PetscCall(MatNullSpaceDestroy(&mat->nearnullsp));
9200: mat->nearnullsp = nullsp;
9201: PetscFunctionReturn(PETSC_SUCCESS);
9202: }
9204: /*@
9205: MatGetNearNullSpace - Get null space attached with `MatSetNearNullSpace()`
9207: Not Collective
9209: Input Parameter:
9210: . mat - the matrix
9212: Output Parameter:
9213: . nullsp - the null space object, `NULL` if not set
9215: Level: advanced
9217: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatNullSpaceCreate()`
9218: @*/
9219: PetscErrorCode MatGetNearNullSpace(Mat mat, MatNullSpace *nullsp)
9220: {
9221: PetscFunctionBegin;
9224: PetscAssertPointer(nullsp, 2);
9225: MatCheckPreallocated(mat, 1);
9226: *nullsp = mat->nearnullsp;
9227: PetscFunctionReturn(PETSC_SUCCESS);
9228: }
9230: /*@
9231: MatICCFactor - Performs in-place incomplete Cholesky factorization of matrix.
9233: Collective
9235: Input Parameters:
9236: + mat - the matrix
9237: . row - row/column permutation
9238: - info - information on desired factorization process
9240: Level: developer
9242: Notes:
9243: Probably really in-place only when level of fill is zero, otherwise allocates
9244: new space to store factored matrix and deletes previous memory.
9246: Most users should employ the `KSP` interface for linear solvers
9247: instead of working directly with matrix algebra routines such as this.
9248: See, e.g., `KSPCreate()`.
9250: Developer Note:
9251: The Fortran interface is not autogenerated as the
9252: interface definition cannot be generated correctly [due to `MatFactorInfo`]
9254: .seealso: [](ch_matrices), `Mat`, `MatFactorInfo`, `MatGetFactor()`, `MatICCFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`
9255: @*/
9256: PetscErrorCode MatICCFactor(Mat mat, IS row, const MatFactorInfo *info)
9257: {
9258: PetscFunctionBegin;
9262: PetscAssertPointer(info, 3);
9263: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "matrix must be square");
9264: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
9265: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
9266: MatCheckPreallocated(mat, 1);
9267: PetscUseTypeMethod(mat, iccfactor, row, info);
9268: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
9269: PetscFunctionReturn(PETSC_SUCCESS);
9270: }
9272: /*@
9273: MatDiagonalScaleLocal - Scales columns of a matrix given the scaling values including the
9274: ghosted ones.
9276: Not Collective
9278: Input Parameters:
9279: + mat - the matrix
9280: - diag - the diagonal values, including ghost ones
9282: Level: developer
9284: Notes:
9285: Works only for `MATMPIAIJ` and `MATMPIBAIJ` matrices
9287: This allows one to avoid during communication to perform the scaling that must be done with `MatDiagonalScale()`
9289: .seealso: [](ch_matrices), `Mat`, `MatDiagonalScale()`
9290: @*/
9291: PetscErrorCode MatDiagonalScaleLocal(Mat mat, Vec diag)
9292: {
9293: PetscMPIInt size;
9295: PetscFunctionBegin;
9300: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Matrix must be already assembled");
9301: PetscCall(PetscLogEventBegin(MAT_Scale, mat, 0, 0, 0));
9302: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
9303: if (size == 1) {
9304: PetscInt n, m;
9305: PetscCall(VecGetSize(diag, &n));
9306: PetscCall(MatGetSize(mat, NULL, &m));
9307: PetscCheck(m == n, PETSC_COMM_SELF, PETSC_ERR_SUP, "Only supported for sequential matrices when no ghost points/periodic conditions");
9308: PetscCall(MatDiagonalScale(mat, NULL, diag));
9309: } else {
9310: PetscUseMethod(mat, "MatDiagonalScaleLocal_C", (Mat, Vec), (mat, diag));
9311: }
9312: PetscCall(PetscLogEventEnd(MAT_Scale, mat, 0, 0, 0));
9313: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
9314: PetscFunctionReturn(PETSC_SUCCESS);
9315: }
9317: /*@
9318: MatGetInertia - Gets the inertia from a factored matrix
9320: Collective
9322: Input Parameter:
9323: . mat - the matrix
9325: Output Parameters:
9326: + nneg - number of negative eigenvalues
9327: . nzero - number of zero eigenvalues
9328: - npos - number of positive eigenvalues
9330: Level: advanced
9332: Note:
9333: Matrix must have been factored by `MatCholeskyFactor()`
9335: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatCholeskyFactor()`
9336: @*/
9337: PetscErrorCode MatGetInertia(Mat mat, PetscInt *nneg, PetscInt *nzero, PetscInt *npos)
9338: {
9339: PetscFunctionBegin;
9342: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Unfactored matrix");
9343: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Numeric factor mat is not assembled");
9344: PetscUseTypeMethod(mat, getinertia, nneg, nzero, npos);
9345: PetscFunctionReturn(PETSC_SUCCESS);
9346: }
9348: /*@C
9349: MatSolves - Solves $A x = b$, given a factored matrix, for a collection of vectors
9351: Neighbor-wise Collective
9353: Input Parameters:
9354: + mat - the factored matrix obtained with `MatGetFactor()`
9355: - b - the right-hand-side vectors
9357: Output Parameter:
9358: . x - the result vectors
9360: Level: developer
9362: Note:
9363: The vectors `b` and `x` cannot be the same. I.e., one cannot
9364: call `MatSolves`(A,x,x).
9366: .seealso: [](ch_matrices), `Mat`, `Vecs`, `MatSolveAdd()`, `MatSolveTranspose()`, `MatSolveTransposeAdd()`, `MatSolve()`
9367: @*/
9368: PetscErrorCode MatSolves(Mat mat, Vecs b, Vecs x)
9369: {
9370: PetscFunctionBegin;
9373: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
9374: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Unfactored matrix");
9375: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
9377: MatCheckPreallocated(mat, 1);
9378: PetscCall(PetscLogEventBegin(MAT_Solves, mat, 0, 0, 0));
9379: PetscUseTypeMethod(mat, solves, b, x);
9380: PetscCall(PetscLogEventEnd(MAT_Solves, mat, 0, 0, 0));
9381: PetscFunctionReturn(PETSC_SUCCESS);
9382: }
9384: /*@
9385: MatIsSymmetric - Test whether a matrix is symmetric
9387: Collective
9389: Input Parameters:
9390: + A - the matrix to test
9391: - tol - difference between value and its transpose less than this amount counts as equal (use 0.0 for exact transpose)
9393: Output Parameter:
9394: . flg - the result
9396: Level: intermediate
9398: Notes:
9399: For real numbers `MatIsSymmetric()` and `MatIsHermitian()` return identical results
9401: If the matrix does not yet know if it is symmetric or not this can be an expensive operation, also available `MatIsSymmetricKnown()`
9403: One can declare that a matrix is symmetric with `MatSetOption`(mat,`MAT_SYMMETRIC`,`PETSC_TRUE`) and if it is known to remain symmetric
9404: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
9406: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetricKnown()`,
9407: `MAT_SYMMETRIC`, `MAT_SYMMETRY_ETERNAL`
9408: @*/
9409: PetscErrorCode MatIsSymmetric(Mat A, PetscReal tol, PetscBool *flg)
9410: {
9411: PetscFunctionBegin;
9413: PetscAssertPointer(flg, 3);
9414: if (A->symmetric != PETSC_BOOL3_UNKNOWN && !tol) *flg = PetscBool3ToBool(A->symmetric);
9415: else {
9416: if (A->ops->issymmetric) PetscUseTypeMethod(A, issymmetric, tol, flg);
9417: else PetscCall(MatIsTranspose(A, A, tol, flg));
9418: if (!tol) PetscCall(MatSetOption(A, MAT_SYMMETRIC, *flg));
9419: }
9420: PetscFunctionReturn(PETSC_SUCCESS);
9421: }
9423: /*@
9424: MatIsHermitian - Test whether a matrix is Hermitian
9426: Collective
9428: Input Parameters:
9429: + A - the matrix to test
9430: - tol - difference between value and its transpose less than this amount counts as equal (use 0.0 for exact Hermitian)
9432: Output Parameter:
9433: . flg - the result
9435: Level: intermediate
9437: Notes:
9438: For real numbers `MatIsSymmetric()` and `MatIsHermitian()` return identical results
9440: If the matrix does not yet know if it is Hermitian or not this can be an expensive operation, also available `MatIsHermitianKnown()`
9442: One can declare that a matrix is Hermitian with `MatSetOption`(mat,`MAT_HERMITIAN`,`PETSC_TRUE`) and if it is known to remain Hermitian
9443: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMEMTRY_ETERNAL`,`PETSC_TRUE`)
9445: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetric()`, `MatSetOption()`,
9446: `MatIsSymmetricKnown()`, `MatIsSymmetric()`, `MAT_HERMITIAN`, `MAT_SYMMETRY_ETERNAL`
9447: @*/
9448: PetscErrorCode MatIsHermitian(Mat A, PetscReal tol, PetscBool *flg)
9449: {
9450: PetscFunctionBegin;
9452: PetscAssertPointer(flg, 3);
9453: if (A->hermitian != PETSC_BOOL3_UNKNOWN && !tol) *flg = PetscBool3ToBool(A->hermitian);
9454: else {
9455: if (A->ops->ishermitian) PetscUseTypeMethod(A, ishermitian, tol, flg);
9456: else PetscCall(MatIsHermitianTranspose(A, A, tol, flg));
9457: if (!tol) PetscCall(MatSetOption(A, MAT_HERMITIAN, *flg));
9458: }
9459: PetscFunctionReturn(PETSC_SUCCESS);
9460: }
9462: /*@
9463: MatIsSymmetricKnown - Checks if a matrix knows if it is symmetric or not and its symmetric state
9465: Not Collective
9467: Input Parameter:
9468: . A - the matrix to check
9470: Output Parameters:
9471: + set - `PETSC_TRUE` if the matrix knows its symmetry state (this tells you if the next flag is valid)
9472: - flg - the result (only valid if set is `PETSC_TRUE`)
9474: Level: advanced
9476: Notes:
9477: Does not check the matrix values directly, so this may return unknown (set = `PETSC_FALSE`). Use `MatIsSymmetric()`
9478: if you want it explicitly checked
9480: One can declare that a matrix is symmetric with `MatSetOption`(mat,`MAT_SYMMETRIC`,`PETSC_TRUE`) and if it is known to remain symmetric
9481: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
9483: .seealso: [](ch_matrices), `Mat`, `MAT_SYMMETRY_ETERNAL`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitianKnown()`
9484: @*/
9485: PetscErrorCode MatIsSymmetricKnown(Mat A, PetscBool *set, PetscBool *flg)
9486: {
9487: PetscFunctionBegin;
9489: PetscAssertPointer(set, 2);
9490: PetscAssertPointer(flg, 3);
9491: if (A->symmetric != PETSC_BOOL3_UNKNOWN) {
9492: *set = PETSC_TRUE;
9493: *flg = PetscBool3ToBool(A->symmetric);
9494: } else {
9495: *set = PETSC_FALSE;
9496: }
9497: PetscFunctionReturn(PETSC_SUCCESS);
9498: }
9500: /*@
9501: MatIsSPDKnown - Checks if a matrix knows if it is symmetric positive definite or not and its symmetric positive definite state
9503: Not Collective
9505: Input Parameter:
9506: . A - the matrix to check
9508: Output Parameters:
9509: + set - `PETSC_TRUE` if the matrix knows its symmetric positive definite state (this tells you if the next flag is valid)
9510: - flg - the result (only valid if set is `PETSC_TRUE`)
9512: Level: advanced
9514: Notes:
9515: Does not check the matrix values directly, so this may return unknown (set = `PETSC_FALSE`).
9517: One can declare that a matrix is SPD with `MatSetOption`(mat,`MAT_SPD`,`PETSC_TRUE`) and if it is known to remain SPD
9518: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SPD_ETERNAL`,`PETSC_TRUE`)
9520: .seealso: [](ch_matrices), `Mat`, `MAT_SPD_ETERNAL`, `MAT_SPD`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitianKnown()`
9521: @*/
9522: PetscErrorCode MatIsSPDKnown(Mat A, PetscBool *set, PetscBool *flg)
9523: {
9524: PetscFunctionBegin;
9526: PetscAssertPointer(set, 2);
9527: PetscAssertPointer(flg, 3);
9528: if (A->spd != PETSC_BOOL3_UNKNOWN) {
9529: *set = PETSC_TRUE;
9530: *flg = PetscBool3ToBool(A->spd);
9531: } else {
9532: *set = PETSC_FALSE;
9533: }
9534: PetscFunctionReturn(PETSC_SUCCESS);
9535: }
9537: /*@
9538: MatIsHermitianKnown - Checks if a matrix knows if it is Hermitian or not and its Hermitian state
9540: Not Collective
9542: Input Parameter:
9543: . A - the matrix to check
9545: Output Parameters:
9546: + set - `PETSC_TRUE` if the matrix knows its Hermitian state (this tells you if the next flag is valid)
9547: - flg - the result (only valid if set is `PETSC_TRUE`)
9549: Level: advanced
9551: Notes:
9552: Does not check the matrix values directly, so this may return unknown (set = `PETSC_FALSE`). Use `MatIsHermitian()`
9553: if you want it explicitly checked
9555: One can declare that a matrix is Hermitian with `MatSetOption`(mat,`MAT_HERMITIAN`,`PETSC_TRUE`) and if it is known to remain Hermitian
9556: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
9558: .seealso: [](ch_matrices), `Mat`, `MAT_SYMMETRY_ETERNAL`, `MAT_HERMITIAN`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`
9559: @*/
9560: PetscErrorCode MatIsHermitianKnown(Mat A, PetscBool *set, PetscBool *flg)
9561: {
9562: PetscFunctionBegin;
9564: PetscAssertPointer(set, 2);
9565: PetscAssertPointer(flg, 3);
9566: if (A->hermitian != PETSC_BOOL3_UNKNOWN) {
9567: *set = PETSC_TRUE;
9568: *flg = PetscBool3ToBool(A->hermitian);
9569: } else {
9570: *set = PETSC_FALSE;
9571: }
9572: PetscFunctionReturn(PETSC_SUCCESS);
9573: }
9575: /*@
9576: MatIsStructurallySymmetric - Test whether a matrix is structurally symmetric
9578: Collective
9580: Input Parameter:
9581: . A - the matrix to test
9583: Output Parameter:
9584: . flg - the result
9586: Level: intermediate
9588: Notes:
9589: If the matrix does yet know it is structurally symmetric this can be an expensive operation, also available `MatIsStructurallySymmetricKnown()`
9591: 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
9592: symmetric after changes to the matrices values one can call `MatSetOption`(mat,`MAT_STRUCTURAL_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
9594: .seealso: [](ch_matrices), `Mat`, `MAT_STRUCTURALLY_SYMMETRIC`, `MAT_STRUCTURAL_SYMMETRY_ETERNAL`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsSymmetric()`, `MatSetOption()`, `MatIsStructurallySymmetricKnown()`
9595: @*/
9596: PetscErrorCode MatIsStructurallySymmetric(Mat A, PetscBool *flg)
9597: {
9598: PetscFunctionBegin;
9600: PetscAssertPointer(flg, 2);
9601: if (A->structurally_symmetric != PETSC_BOOL3_UNKNOWN) {
9602: *flg = PetscBool3ToBool(A->structurally_symmetric);
9603: } else {
9604: PetscUseTypeMethod(A, isstructurallysymmetric, flg);
9605: PetscCall(MatSetOption(A, MAT_STRUCTURALLY_SYMMETRIC, *flg));
9606: }
9607: PetscFunctionReturn(PETSC_SUCCESS);
9608: }
9610: /*@
9611: MatIsStructurallySymmetricKnown - Checks if a matrix knows if it is structurally symmetric or not and its structurally symmetric state
9613: Not Collective
9615: Input Parameter:
9616: . A - the matrix to check
9618: Output Parameters:
9619: + set - PETSC_TRUE if the matrix knows its structurally symmetric state (this tells you if the next flag is valid)
9620: - flg - the result (only valid if set is PETSC_TRUE)
9622: Level: advanced
9624: Notes:
9625: 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
9626: symmetric after changes to the matrices values one can call `MatSetOption`(mat,`MAT_STRUCTURAL_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
9628: Use `MatIsStructurallySymmetric()` to explicitly check if a matrix is structurally symmetric (this is an expensive operation)
9630: .seealso: [](ch_matrices), `Mat`, `MAT_STRUCTURALLY_SYMMETRIC`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitianKnown()`
9631: @*/
9632: PetscErrorCode MatIsStructurallySymmetricKnown(Mat A, PetscBool *set, PetscBool *flg)
9633: {
9634: PetscFunctionBegin;
9636: PetscAssertPointer(set, 2);
9637: PetscAssertPointer(flg, 3);
9638: if (A->structurally_symmetric != PETSC_BOOL3_UNKNOWN) {
9639: *set = PETSC_TRUE;
9640: *flg = PetscBool3ToBool(A->structurally_symmetric);
9641: } else {
9642: *set = PETSC_FALSE;
9643: }
9644: PetscFunctionReturn(PETSC_SUCCESS);
9645: }
9647: /*@
9648: MatStashGetInfo - Gets how many values are currently in the matrix stash, i.e. need
9649: to be communicated to other processors during the `MatAssemblyBegin()`/`MatAssemblyEnd()` process
9651: Not Collective
9653: Input Parameter:
9654: . mat - the matrix
9656: Output Parameters:
9657: + nstash - the size of the stash
9658: . reallocs - the number of additional mallocs incurred.
9659: . bnstash - the size of the block stash
9660: - breallocs - the number of additional mallocs incurred.in the block stash
9662: Level: advanced
9664: .seealso: [](ch_matrices), `MatAssemblyBegin()`, `MatAssemblyEnd()`, `Mat`, `MatStashSetInitialSize()`
9665: @*/
9666: PetscErrorCode MatStashGetInfo(Mat mat, PetscInt *nstash, PetscInt *reallocs, PetscInt *bnstash, PetscInt *breallocs)
9667: {
9668: PetscFunctionBegin;
9669: PetscCall(MatStashGetInfo_Private(&mat->stash, nstash, reallocs));
9670: PetscCall(MatStashGetInfo_Private(&mat->bstash, bnstash, breallocs));
9671: PetscFunctionReturn(PETSC_SUCCESS);
9672: }
9674: /*@
9675: MatCreateVecs - Get vector(s) compatible with the matrix, i.e. with the same
9676: parallel layout, `PetscLayout` for rows and columns
9678: Collective
9680: Input Parameter:
9681: . mat - the matrix
9683: Output Parameters:
9684: + right - (optional) vector that the matrix can be multiplied against
9685: - left - (optional) vector that the matrix vector product can be stored in
9687: Level: advanced
9689: Notes:
9690: 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()`.
9692: These are new vectors which are not owned by the mat, they should be destroyed in `VecDestroy()` when no longer needed
9694: .seealso: [](ch_matrices), `Mat`, `Vec`, `VecCreate()`, `VecDestroy()`, `DMCreateGlobalVector()`
9695: @*/
9696: PetscErrorCode MatCreateVecs(Mat mat, Vec *right, Vec *left)
9697: {
9698: PetscFunctionBegin;
9701: if (mat->ops->getvecs) {
9702: PetscUseTypeMethod(mat, getvecs, right, left);
9703: } else {
9704: if (right) {
9705: PetscCheck(mat->cmap->n >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "PetscLayout for columns not yet setup");
9706: PetscCall(VecCreateWithLayout_Private(mat->cmap, right));
9707: PetscCall(VecSetType(*right, mat->defaultvectype));
9708: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_HIP)
9709: if (mat->boundtocpu && mat->bindingpropagates) {
9710: PetscCall(VecSetBindingPropagates(*right, PETSC_TRUE));
9711: PetscCall(VecBindToCPU(*right, PETSC_TRUE));
9712: }
9713: #endif
9714: }
9715: if (left) {
9716: PetscCheck(mat->rmap->n >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "PetscLayout for rows not yet setup");
9717: PetscCall(VecCreateWithLayout_Private(mat->rmap, left));
9718: PetscCall(VecSetType(*left, mat->defaultvectype));
9719: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_HIP)
9720: if (mat->boundtocpu && mat->bindingpropagates) {
9721: PetscCall(VecSetBindingPropagates(*left, PETSC_TRUE));
9722: PetscCall(VecBindToCPU(*left, PETSC_TRUE));
9723: }
9724: #endif
9725: }
9726: }
9727: PetscFunctionReturn(PETSC_SUCCESS);
9728: }
9730: /*@
9731: MatFactorInfoInitialize - Initializes a `MatFactorInfo` data structure
9732: with default values.
9734: Not Collective
9736: Input Parameter:
9737: . info - the `MatFactorInfo` data structure
9739: Level: developer
9741: Notes:
9742: The solvers are generally used through the `KSP` and `PC` objects, for example
9743: `PCLU`, `PCILU`, `PCCHOLESKY`, `PCICC`
9745: Once the data structure is initialized one may change certain entries as desired for the particular factorization to be performed
9747: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorInfo`
9748: @*/
9749: PetscErrorCode MatFactorInfoInitialize(MatFactorInfo *info)
9750: {
9751: PetscFunctionBegin;
9752: PetscCall(PetscMemzero(info, sizeof(MatFactorInfo)));
9753: PetscFunctionReturn(PETSC_SUCCESS);
9754: }
9756: /*@
9757: MatFactorSetSchurIS - Set indices corresponding to the Schur complement you wish to have computed
9759: Collective
9761: Input Parameters:
9762: + mat - the factored matrix
9763: - is - the index set defining the Schur indices (0-based)
9765: Level: advanced
9767: Notes:
9768: Call `MatFactorSolveSchurComplement()` or `MatFactorSolveSchurComplementTranspose()` after this call to solve a Schur complement system.
9770: You can call `MatFactorGetSchurComplement()` or `MatFactorCreateSchurComplement()` after this call.
9772: This functionality is only supported for `MATSOLVERMUMPS` and `MATSOLVERMKL_PARDISO`
9774: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorGetSchurComplement()`, `MatFactorRestoreSchurComplement()`, `MatFactorCreateSchurComplement()`, `MatFactorSolveSchurComplement()`,
9775: `MatFactorSolveSchurComplementTranspose()`, `MATSOLVERMUMPS`, `MATSOLVERMKL_PARDISO`
9776: @*/
9777: PetscErrorCode MatFactorSetSchurIS(Mat mat, IS is)
9778: {
9779: PetscErrorCode (*f)(Mat, IS);
9781: PetscFunctionBegin;
9786: PetscCheckSameComm(mat, 1, is, 2);
9787: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Only for factored matrix");
9788: PetscCall(PetscObjectQueryFunction((PetscObject)mat, "MatFactorSetSchurIS_C", &f));
9789: PetscCheck(f, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "The selected MatSolverType does not support Schur complement computation. You should use MATSOLVERMUMPS or MATSOLVERMKL_PARDISO");
9790: PetscCall(MatDestroy(&mat->schur));
9791: PetscCall((*f)(mat, is));
9792: PetscCheck(mat->schur, PetscObjectComm((PetscObject)mat), PETSC_ERR_PLIB, "Schur complement has not been created");
9793: PetscFunctionReturn(PETSC_SUCCESS);
9794: }
9796: /*@
9797: MatFactorCreateSchurComplement - Create a Schur complement matrix object using Schur data computed during the factorization step
9799: Logically Collective
9801: Input Parameters:
9802: + F - the factored matrix obtained by calling `MatGetFactor()`
9803: . S - location where to return the Schur complement, can be `NULL`
9804: - status - the status of the Schur complement matrix, can be `NULL`
9806: Level: advanced
9808: Notes:
9809: You must call `MatFactorSetSchurIS()` before calling this routine.
9811: This functionality is only supported for `MATSOLVERMUMPS` and `MATSOLVERMKL_PARDISO`
9813: The routine provides a copy of the Schur matrix stored within the solver data structures.
9814: The caller must destroy the object when it is no longer needed.
9815: If `MatFactorInvertSchurComplement()` has been called, the routine gets back the inverse.
9817: 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)
9819: See `MatCreateSchurComplement()` or `MatGetSchurComplement()` for ways to create virtual or approximate Schur complements.
9821: Developer Note:
9822: The reason this routine exists is because the representation of the Schur complement within the factor matrix may be different than a standard PETSc
9823: matrix representation and we normally do not want to use the time or memory to make a copy as a regular PETSc matrix.
9825: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorGetSchurComplement()`, `MatFactorSchurStatus`, `MATSOLVERMUMPS`, `MATSOLVERMKL_PARDISO`
9826: @*/
9827: PetscErrorCode MatFactorCreateSchurComplement(Mat F, Mat *S, MatFactorSchurStatus *status)
9828: {
9829: PetscFunctionBegin;
9831: if (S) PetscAssertPointer(S, 2);
9832: if (status) PetscAssertPointer(status, 3);
9833: if (S) {
9834: PetscErrorCode (*f)(Mat, Mat *);
9836: PetscCall(PetscObjectQueryFunction((PetscObject)F, "MatFactorCreateSchurComplement_C", &f));
9837: if (f) {
9838: PetscCall((*f)(F, S));
9839: } else {
9840: PetscCall(MatDuplicate(F->schur, MAT_COPY_VALUES, S));
9841: }
9842: }
9843: if (status) *status = F->schur_status;
9844: PetscFunctionReturn(PETSC_SUCCESS);
9845: }
9847: /*@
9848: MatFactorGetSchurComplement - Gets access to a Schur complement matrix using the current Schur data within a factored matrix
9850: Logically Collective
9852: Input Parameters:
9853: + F - the factored matrix obtained by calling `MatGetFactor()`
9854: . S - location where to return the Schur complement, can be `NULL`
9855: - status - the status of the Schur complement matrix, can be `NULL`
9857: Level: advanced
9859: Notes:
9860: You must call `MatFactorSetSchurIS()` before calling this routine.
9862: Schur complement mode is currently implemented for sequential matrices with factor type of `MATSOLVERMUMPS`
9864: The routine returns a the Schur Complement stored within the data structures of the solver.
9866: If `MatFactorInvertSchurComplement()` has previously been called, the returned matrix is actually the inverse of the Schur complement.
9868: The returned matrix should not be destroyed; the caller should call `MatFactorRestoreSchurComplement()` when the object is no longer needed.
9870: Use `MatFactorCreateSchurComplement()` to create a copy of the Schur complement matrix that is within a factored matrix
9872: See `MatCreateSchurComplement()` or `MatGetSchurComplement()` for ways to create virtual or approximate Schur complements.
9874: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorRestoreSchurComplement()`, `MatFactorCreateSchurComplement()`, `MatFactorSchurStatus`
9875: @*/
9876: PetscErrorCode MatFactorGetSchurComplement(Mat F, Mat *S, MatFactorSchurStatus *status)
9877: {
9878: PetscFunctionBegin;
9880: if (S) {
9881: PetscAssertPointer(S, 2);
9882: *S = F->schur;
9883: }
9884: if (status) {
9885: PetscAssertPointer(status, 3);
9886: *status = F->schur_status;
9887: }
9888: PetscFunctionReturn(PETSC_SUCCESS);
9889: }
9891: static PetscErrorCode MatFactorUpdateSchurStatus_Private(Mat F)
9892: {
9893: Mat S = F->schur;
9895: PetscFunctionBegin;
9896: switch (F->schur_status) {
9897: case MAT_FACTOR_SCHUR_UNFACTORED: // fall-through
9898: case MAT_FACTOR_SCHUR_INVERTED:
9899: if (S) {
9900: S->ops->solve = NULL;
9901: S->ops->matsolve = NULL;
9902: S->ops->solvetranspose = NULL;
9903: S->ops->matsolvetranspose = NULL;
9904: S->ops->solveadd = NULL;
9905: S->ops->solvetransposeadd = NULL;
9906: S->factortype = MAT_FACTOR_NONE;
9907: PetscCall(PetscFree(S->solvertype));
9908: }
9909: case MAT_FACTOR_SCHUR_FACTORED: // fall-through
9910: break;
9911: default:
9912: SETERRQ(PetscObjectComm((PetscObject)F), PETSC_ERR_SUP, "Unhandled MatFactorSchurStatus %d", F->schur_status);
9913: }
9914: PetscFunctionReturn(PETSC_SUCCESS);
9915: }
9917: /*@
9918: MatFactorRestoreSchurComplement - Restore the Schur complement matrix object obtained from a call to `MatFactorGetSchurComplement()`
9920: Logically Collective
9922: Input Parameters:
9923: + F - the factored matrix obtained by calling `MatGetFactor()`
9924: . S - location where the Schur complement is stored
9925: - status - the status of the Schur complement matrix (see `MatFactorSchurStatus`)
9927: Level: advanced
9929: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorCreateSchurComplement()`, `MatFactorSchurStatus`
9930: @*/
9931: PetscErrorCode MatFactorRestoreSchurComplement(Mat F, Mat *S, MatFactorSchurStatus status)
9932: {
9933: PetscFunctionBegin;
9935: if (S) {
9937: *S = NULL;
9938: }
9939: F->schur_status = status;
9940: PetscCall(MatFactorUpdateSchurStatus_Private(F));
9941: PetscFunctionReturn(PETSC_SUCCESS);
9942: }
9944: /*@
9945: MatFactorSolveSchurComplementTranspose - Solve the transpose of the Schur complement system computed during the factorization step
9947: Logically Collective
9949: Input Parameters:
9950: + F - the factored matrix obtained by calling `MatGetFactor()`
9951: . rhs - location where the right-hand side of the Schur complement system is stored
9952: - sol - location where the solution of the Schur complement system has to be returned
9954: Level: advanced
9956: Notes:
9957: The sizes of the vectors should match the size of the Schur complement
9959: Must be called after `MatFactorSetSchurIS()`
9961: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorSolveSchurComplement()`
9962: @*/
9963: PetscErrorCode MatFactorSolveSchurComplementTranspose(Mat F, Vec rhs, Vec sol)
9964: {
9965: PetscFunctionBegin;
9972: PetscCheckSameComm(F, 1, rhs, 2);
9973: PetscCheckSameComm(F, 1, sol, 3);
9974: PetscCall(MatFactorFactorizeSchurComplement(F));
9975: switch (F->schur_status) {
9976: case MAT_FACTOR_SCHUR_FACTORED:
9977: PetscCall(MatSolveTranspose(F->schur, rhs, sol));
9978: break;
9979: case MAT_FACTOR_SCHUR_INVERTED:
9980: PetscCall(MatMultTranspose(F->schur, rhs, sol));
9981: break;
9982: default:
9983: SETERRQ(PetscObjectComm((PetscObject)F), PETSC_ERR_SUP, "Unhandled MatFactorSchurStatus %d", F->schur_status);
9984: }
9985: PetscFunctionReturn(PETSC_SUCCESS);
9986: }
9988: /*@
9989: MatFactorSolveSchurComplement - Solve the Schur complement system computed during the factorization step
9991: Logically Collective
9993: Input Parameters:
9994: + F - the factored matrix obtained by calling `MatGetFactor()`
9995: . rhs - location where the right-hand side of the Schur complement system is stored
9996: - sol - location where the solution of the Schur complement system has to be returned
9998: Level: advanced
10000: Notes:
10001: The sizes of the vectors should match the size of the Schur complement
10003: Must be called after `MatFactorSetSchurIS()`
10005: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorSolveSchurComplementTranspose()`
10006: @*/
10007: PetscErrorCode MatFactorSolveSchurComplement(Mat F, Vec rhs, Vec sol)
10008: {
10009: PetscFunctionBegin;
10016: PetscCheckSameComm(F, 1, rhs, 2);
10017: PetscCheckSameComm(F, 1, sol, 3);
10018: PetscCall(MatFactorFactorizeSchurComplement(F));
10019: switch (F->schur_status) {
10020: case MAT_FACTOR_SCHUR_FACTORED:
10021: PetscCall(MatSolve(F->schur, rhs, sol));
10022: break;
10023: case MAT_FACTOR_SCHUR_INVERTED:
10024: PetscCall(MatMult(F->schur, rhs, sol));
10025: break;
10026: default:
10027: SETERRQ(PetscObjectComm((PetscObject)F), PETSC_ERR_SUP, "Unhandled MatFactorSchurStatus %d", F->schur_status);
10028: }
10029: PetscFunctionReturn(PETSC_SUCCESS);
10030: }
10032: PETSC_EXTERN PetscErrorCode MatSeqDenseInvertFactors_Private(Mat);
10033: #if PetscDefined(HAVE_CUDA)
10034: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatSeqDenseCUDAInvertFactors_Internal(Mat);
10035: #endif
10037: /* Schur status updated in the interface */
10038: static PetscErrorCode MatFactorInvertSchurComplement_Private(Mat F)
10039: {
10040: Mat S = F->schur;
10042: PetscFunctionBegin;
10043: if (S) {
10044: PetscMPIInt size;
10045: PetscBool isdense, isdensecuda;
10047: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)S), &size));
10048: PetscCheck(size <= 1, PetscObjectComm((PetscObject)S), PETSC_ERR_SUP, "Not yet implemented");
10049: PetscCall(PetscObjectTypeCompare((PetscObject)S, MATSEQDENSE, &isdense));
10050: PetscCall(PetscObjectTypeCompare((PetscObject)S, MATSEQDENSECUDA, &isdensecuda));
10051: PetscCheck(isdense || isdensecuda, PetscObjectComm((PetscObject)S), PETSC_ERR_SUP, "Not implemented for type %s", ((PetscObject)S)->type_name);
10052: PetscCall(PetscLogEventBegin(MAT_FactorInvS, F, 0, 0, 0));
10053: if (isdense) {
10054: PetscCall(MatSeqDenseInvertFactors_Private(S));
10055: } else if (isdensecuda) {
10056: #if defined(PETSC_HAVE_CUDA)
10057: PetscCall(MatSeqDenseCUDAInvertFactors_Internal(S));
10058: #endif
10059: }
10060: // HIP??????????????
10061: PetscCall(PetscLogEventEnd(MAT_FactorInvS, F, 0, 0, 0));
10062: }
10063: PetscFunctionReturn(PETSC_SUCCESS);
10064: }
10066: /*@
10067: MatFactorInvertSchurComplement - Invert the Schur complement matrix computed during the factorization step
10069: Logically Collective
10071: Input Parameter:
10072: . F - the factored matrix obtained by calling `MatGetFactor()`
10074: Level: advanced
10076: Notes:
10077: Must be called after `MatFactorSetSchurIS()`.
10079: Call `MatFactorGetSchurComplement()` or `MatFactorCreateSchurComplement()` AFTER this call to actually compute the inverse and get access to it.
10081: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorGetSchurComplement()`, `MatFactorCreateSchurComplement()`
10082: @*/
10083: PetscErrorCode MatFactorInvertSchurComplement(Mat F)
10084: {
10085: PetscFunctionBegin;
10088: if (F->schur_status == MAT_FACTOR_SCHUR_INVERTED) PetscFunctionReturn(PETSC_SUCCESS);
10089: PetscCall(MatFactorFactorizeSchurComplement(F));
10090: PetscCall(MatFactorInvertSchurComplement_Private(F));
10091: F->schur_status = MAT_FACTOR_SCHUR_INVERTED;
10092: PetscFunctionReturn(PETSC_SUCCESS);
10093: }
10095: /*@
10096: MatFactorFactorizeSchurComplement - Factorize the Schur complement matrix computed during the factorization step
10098: Logically Collective
10100: Input Parameter:
10101: . F - the factored matrix obtained by calling `MatGetFactor()`
10103: Level: advanced
10105: Note:
10106: Must be called after `MatFactorSetSchurIS()`
10108: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorInvertSchurComplement()`
10109: @*/
10110: PetscErrorCode MatFactorFactorizeSchurComplement(Mat F)
10111: {
10112: MatFactorInfo info;
10114: PetscFunctionBegin;
10117: if (F->schur_status == MAT_FACTOR_SCHUR_INVERTED || F->schur_status == MAT_FACTOR_SCHUR_FACTORED) PetscFunctionReturn(PETSC_SUCCESS);
10118: PetscCall(PetscLogEventBegin(MAT_FactorFactS, F, 0, 0, 0));
10119: PetscCall(PetscMemzero(&info, sizeof(MatFactorInfo)));
10120: if (F->factortype == MAT_FACTOR_CHOLESKY) { /* LDL^t regarded as Cholesky */
10121: PetscCall(MatCholeskyFactor(F->schur, NULL, &info));
10122: } else {
10123: PetscCall(MatLUFactor(F->schur, NULL, NULL, &info));
10124: }
10125: PetscCall(PetscLogEventEnd(MAT_FactorFactS, F, 0, 0, 0));
10126: F->schur_status = MAT_FACTOR_SCHUR_FACTORED;
10127: PetscFunctionReturn(PETSC_SUCCESS);
10128: }
10130: /*@
10131: MatPtAP - Creates the matrix product $C = P^T * A * P$
10133: Neighbor-wise Collective
10135: Input Parameters:
10136: + A - the matrix
10137: . P - the projection matrix
10138: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10139: - 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
10140: if the result is a dense matrix this is irrelevant
10142: Output Parameter:
10143: . C - the product matrix
10145: Level: intermediate
10147: Notes:
10148: C will be created and must be destroyed by the user with `MatDestroy()`.
10150: An alternative approach to this function is to use `MatProductCreate()` and set the desired options before the computation is done
10152: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10154: Developer Note:
10155: For matrix types without special implementation the function fallbacks to `MatMatMult()` followed by `MatTransposeMatMult()`.
10157: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MatMatMult()`, `MatRARt()`
10158: @*/
10159: PetscErrorCode MatPtAP(Mat A, Mat P, MatReuse scall, PetscReal fill, Mat *C)
10160: {
10161: PetscFunctionBegin;
10162: if (scall == MAT_REUSE_MATRIX) MatCheckProduct(*C, 5);
10163: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
10165: if (scall == MAT_INITIAL_MATRIX) {
10166: PetscCall(MatProductCreate(A, P, NULL, C));
10167: PetscCall(MatProductSetType(*C, MATPRODUCT_PtAP));
10168: PetscCall(MatProductSetAlgorithm(*C, "default"));
10169: PetscCall(MatProductSetFill(*C, fill));
10171: (*C)->product->api_user = PETSC_TRUE;
10172: PetscCall(MatProductSetFromOptions(*C));
10173: 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);
10174: PetscCall(MatProductSymbolic(*C));
10175: } else { /* scall == MAT_REUSE_MATRIX */
10176: PetscCall(MatProductReplaceMats(A, P, NULL, *C));
10177: }
10179: PetscCall(MatProductNumeric(*C));
10180: (*C)->symmetric = A->symmetric;
10181: (*C)->spd = A->spd;
10182: PetscFunctionReturn(PETSC_SUCCESS);
10183: }
10185: /*@
10186: MatRARt - Creates the matrix product $C = R * A * R^T$
10188: Neighbor-wise Collective
10190: Input Parameters:
10191: + A - the matrix
10192: . R - the projection matrix
10193: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10194: - fill - expected fill as ratio of nnz(C)/nnz(A), use `PETSC_DETERMINE` or `PETSC_CURRENT` if you do not have a good estimate
10195: if the result is a dense matrix this is irrelevant
10197: Output Parameter:
10198: . C - the product matrix
10200: Level: intermediate
10202: Notes:
10203: `C` will be created and must be destroyed by the user with `MatDestroy()`.
10205: An alternative approach to this function is to use `MatProductCreate()` and set the desired options before the computation is done
10207: This routine is currently only implemented for pairs of `MATAIJ` matrices and classes
10208: which inherit from `MATAIJ`. Due to PETSc sparse matrix block row distribution among processes,
10209: the parallel `MatRARt()` is implemented computing the explicit transpose of `R`, which can be very expensive.
10210: We recommend using `MatPtAP()` when possible.
10212: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10214: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MatMatMult()`, `MatPtAP()`
10215: @*/
10216: PetscErrorCode MatRARt(Mat A, Mat R, MatReuse scall, PetscReal fill, Mat *C)
10217: {
10218: PetscFunctionBegin;
10219: if (scall == MAT_REUSE_MATRIX) MatCheckProduct(*C, 5);
10220: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
10222: if (scall == MAT_INITIAL_MATRIX) {
10223: PetscCall(MatProductCreate(A, R, NULL, C));
10224: PetscCall(MatProductSetType(*C, MATPRODUCT_RARt));
10225: PetscCall(MatProductSetAlgorithm(*C, "default"));
10226: PetscCall(MatProductSetFill(*C, fill));
10228: (*C)->product->api_user = PETSC_TRUE;
10229: PetscCall(MatProductSetFromOptions(*C));
10230: 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);
10231: PetscCall(MatProductSymbolic(*C));
10232: } else { /* scall == MAT_REUSE_MATRIX */
10233: PetscCall(MatProductReplaceMats(A, R, NULL, *C));
10234: }
10236: PetscCall(MatProductNumeric(*C));
10237: if (A->symmetric == PETSC_BOOL3_TRUE) PetscCall(MatSetOption(*C, MAT_SYMMETRIC, PETSC_TRUE));
10238: PetscFunctionReturn(PETSC_SUCCESS);
10239: }
10241: static PetscErrorCode MatProduct_Private(Mat A, Mat B, MatReuse scall, PetscReal fill, MatProductType ptype, Mat *C)
10242: {
10243: PetscBool flg = PETSC_TRUE;
10245: PetscFunctionBegin;
10246: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "MAT_INPLACE_MATRIX product not supported");
10247: if (scall == MAT_INITIAL_MATRIX) {
10248: PetscCall(PetscInfo(A, "Calling MatProduct API with MAT_INITIAL_MATRIX and product type %s\n", MatProductTypes[ptype]));
10249: PetscCall(MatProductCreate(A, B, NULL, C));
10250: PetscCall(MatProductSetAlgorithm(*C, MATPRODUCTALGORITHMDEFAULT));
10251: PetscCall(MatProductSetFill(*C, fill));
10252: } else { /* scall == MAT_REUSE_MATRIX */
10253: Mat_Product *product = (*C)->product;
10255: PetscCall(PetscObjectBaseTypeCompareAny((PetscObject)*C, &flg, MATSEQDENSE, MATMPIDENSE, ""));
10256: if (flg && product && product->type != ptype) {
10257: PetscCall(MatProductClear(*C));
10258: product = NULL;
10259: }
10260: PetscCall(PetscInfo(A, "Calling MatProduct API with MAT_REUSE_MATRIX %s product present and product type %s\n", product ? "with" : "without", MatProductTypes[ptype]));
10261: if (!product) { /* user provide the dense matrix *C without calling MatProductCreate() or reusing it from previous calls */
10262: PetscCheck(flg, PetscObjectComm((PetscObject)*C), PETSC_ERR_SUP, "Call MatProductCreate() first");
10263: PetscCall(MatProductCreate_Private(A, B, NULL, *C));
10264: product = (*C)->product;
10265: product->fill = fill;
10266: product->clear = PETSC_TRUE;
10267: } else { /* user may change input matrices A or B when MAT_REUSE_MATRIX */
10268: flg = PETSC_FALSE;
10269: PetscCall(MatProductReplaceMats(A, B, NULL, *C));
10270: }
10271: }
10272: if (flg) {
10273: (*C)->product->api_user = PETSC_TRUE;
10274: PetscCall(MatProductSetType(*C, ptype));
10275: PetscCall(MatProductSetFromOptions(*C));
10276: PetscCall(MatProductSymbolic(*C));
10277: }
10278: PetscCall(MatProductNumeric(*C));
10279: PetscFunctionReturn(PETSC_SUCCESS);
10280: }
10282: /*@
10283: MatMatMult - Performs matrix-matrix multiplication C=A*B.
10285: Neighbor-wise Collective
10287: Input Parameters:
10288: + A - the left matrix
10289: . B - the right matrix
10290: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10291: - 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
10292: if the result is a dense matrix this is irrelevant
10294: Output Parameter:
10295: . C - the product matrix
10297: Notes:
10298: Unless scall is `MAT_REUSE_MATRIX` C will be created.
10300: `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
10301: call to this function with `MAT_INITIAL_MATRIX`.
10303: To determine the correct fill value, run with `-info` and search for the string "Fill ratio" to see the value actually needed.
10305: 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`,
10306: rather than first having `MatMatMult()` create it for you. You can NEVER do this if the matrix `C` is sparse.
10308: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10310: Example of Usage:
10311: .vb
10312: MatProductCreate(A,B,NULL,&C);
10313: MatProductSetType(C,MATPRODUCT_AB);
10314: MatProductSymbolic(C);
10315: MatProductNumeric(C); // compute C=A * B
10316: MatProductReplaceMats(A1,B1,NULL,C); // compute C=A1 * B1
10317: MatProductNumeric(C);
10318: MatProductReplaceMats(A2,NULL,NULL,C); // compute C=A2 * B1
10319: MatProductNumeric(C);
10320: .ve
10322: Level: intermediate
10324: .seealso: [](ch_matrices), `Mat`, `MatProductType`, `MATPRODUCT_AB`, `MatTransposeMatMult()`, `MatMatTransposeMult()`, `MatPtAP()`, `MatProductCreate()`, `MatProductSymbolic()`, `MatProductReplaceMats()`, `MatProductNumeric()`
10325: @*/
10326: PetscErrorCode MatMatMult(Mat A, Mat B, MatReuse scall, PetscReal fill, Mat *C)
10327: {
10328: PetscFunctionBegin;
10329: PetscCall(MatProduct_Private(A, B, scall, fill, MATPRODUCT_AB, C));
10330: PetscFunctionReturn(PETSC_SUCCESS);
10331: }
10333: /*@
10334: MatMatTransposeMult - Performs matrix-matrix multiplication $C = A*B^T$.
10336: Neighbor-wise Collective
10338: Input Parameters:
10339: + A - the left matrix
10340: . B - the right matrix
10341: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10342: - fill - expected fill as ratio of nnz(C)/(nnz(A) + nnz(B)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if not known
10344: Output Parameter:
10345: . C - the product matrix
10347: Options Database Key:
10348: . -matmattransmult_mpidense_mpidense_via {allgatherv,cyclic} - Choose between algorithms for `MATMPIDENSE` matrices: the
10349: first redundantly copies the transposed `B` matrix on each process and requires O(log P) communication complexity;
10350: the second never stores more than one portion of the `B` matrix at a time but requires O(P) communication complexity.
10352: Level: intermediate
10354: Notes:
10355: C will be created if `MAT_INITIAL_MATRIX` and must be destroyed by the user with `MatDestroy()`.
10357: `MAT_REUSE_MATRIX` can only be used if the matrices A and B have the same nonzero pattern as in the previous call
10359: To determine the correct fill value, run with -info and search for the string "Fill ratio" to see the value
10360: actually needed.
10362: This routine is currently only implemented for pairs of `MATSEQAIJ` matrices, for the `MATSEQDENSE` class,
10363: and for pairs of `MATMPIDENSE` matrices.
10365: This routine is shorthand for using `MatProductCreate()` with the `MatProductType` of `MATPRODUCT_ABt`
10367: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10369: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MATPRODUCT_ABt`, `MatMatMult()`, `MatTransposeMatMult()` `MatPtAP()`, `MatProductAlgorithm`, `MatProductType`
10370: @*/
10371: PetscErrorCode MatMatTransposeMult(Mat A, Mat B, MatReuse scall, PetscReal fill, Mat *C)
10372: {
10373: PetscFunctionBegin;
10374: PetscCall(MatProduct_Private(A, B, scall, fill, MATPRODUCT_ABt, C));
10375: if (A == B) PetscCall(MatSetOption(*C, MAT_SYMMETRIC, PETSC_TRUE));
10376: PetscFunctionReturn(PETSC_SUCCESS);
10377: }
10379: /*@
10380: MatTransposeMatMult - Performs matrix-matrix multiplication $C = A^T*B$.
10382: Neighbor-wise Collective
10384: Input Parameters:
10385: + A - the left matrix
10386: . B - the right matrix
10387: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10388: - fill - expected fill as ratio of nnz(C)/(nnz(A) + nnz(B)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if not known
10390: Output Parameter:
10391: . C - the product matrix
10393: Level: intermediate
10395: Notes:
10396: `C` will be created if `MAT_INITIAL_MATRIX` and must be destroyed by the user with `MatDestroy()`.
10398: `MAT_REUSE_MATRIX` can only be used if the matrices A and B have the same nonzero pattern as in the previous call.
10400: This routine is shorthand for using `MatProductCreate()` with the `MatProductType` of `MATPRODUCT_AtB`
10402: To determine the correct fill value, run with -info and search for the string "Fill ratio" to see the value
10403: actually needed.
10405: This routine is currently implemented for pairs of `MATAIJ` matrices and pairs of `MATSEQDENSE` matrices and classes
10406: which inherit from `MATSEQAIJ`. `C` will be of the same type as the input matrices.
10408: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10410: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MATPRODUCT_AtB`, `MatMatMult()`, `MatMatTransposeMult()`, `MatPtAP()`
10411: @*/
10412: PetscErrorCode MatTransposeMatMult(Mat A, Mat B, MatReuse scall, PetscReal fill, Mat *C)
10413: {
10414: PetscFunctionBegin;
10415: PetscCall(MatProduct_Private(A, B, scall, fill, MATPRODUCT_AtB, C));
10416: PetscFunctionReturn(PETSC_SUCCESS);
10417: }
10419: /*@
10420: MatMatMatMult - Performs matrix-matrix-matrix multiplication D=A*B*C.
10422: Neighbor-wise Collective
10424: Input Parameters:
10425: + A - the left matrix
10426: . B - the middle matrix
10427: . C - the right matrix
10428: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10429: - 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
10430: if the result is a dense matrix this is irrelevant
10432: Output Parameter:
10433: . D - the product matrix
10435: Level: intermediate
10437: Notes:
10438: Unless `scall` is `MAT_REUSE_MATRIX` `D` will be created.
10440: `MAT_REUSE_MATRIX` can only be used if the matrices `A`, `B`, and `C` have the same nonzero pattern as in the previous call
10442: This routine is shorthand for using `MatProductCreate()` with the `MatProductType` of `MATPRODUCT_ABC`
10444: To determine the correct fill value, run with `-info` and search for the string "Fill ratio" to see the value
10445: actually needed.
10447: If you have many matrices with the same non-zero structure to multiply, you
10448: should use `MAT_REUSE_MATRIX` in all calls but the first
10450: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10452: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MATPRODUCT_ABC`, `MatMatMult`, `MatPtAP()`, `MatMatTransposeMult()`, `MatTransposeMatMult()`
10453: @*/
10454: PetscErrorCode MatMatMatMult(Mat A, Mat B, Mat C, MatReuse scall, PetscReal fill, Mat *D)
10455: {
10456: PetscFunctionBegin;
10457: if (scall == MAT_REUSE_MATRIX) MatCheckProduct(*D, 6);
10458: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
10460: if (scall == MAT_INITIAL_MATRIX) {
10461: PetscCall(MatProductCreate(A, B, C, D));
10462: PetscCall(MatProductSetType(*D, MATPRODUCT_ABC));
10463: PetscCall(MatProductSetAlgorithm(*D, "default"));
10464: PetscCall(MatProductSetFill(*D, fill));
10466: (*D)->product->api_user = PETSC_TRUE;
10467: PetscCall(MatProductSetFromOptions(*D));
10468: 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,
10469: ((PetscObject)C)->type_name);
10470: PetscCall(MatProductSymbolic(*D));
10471: } else { /* user may change input matrices when REUSE */
10472: PetscCall(MatProductReplaceMats(A, B, C, *D));
10473: }
10474: PetscCall(MatProductNumeric(*D));
10475: PetscFunctionReturn(PETSC_SUCCESS);
10476: }
10478: /*@
10479: MatCreateRedundantMatrix - Create redundant matrices and put them into processors of subcommunicators.
10481: Collective
10483: Input Parameters:
10484: + mat - the matrix
10485: . nsubcomm - the number of subcommunicators (= number of redundant parallel or sequential matrices)
10486: . subcomm - MPI communicator split from the communicator where mat resides in (or `MPI_COMM_NULL` if nsubcomm is used)
10487: - reuse - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10489: Output Parameter:
10490: . matredundant - redundant matrix
10492: Level: advanced
10494: Notes:
10495: `MAT_REUSE_MATRIX` can only be used when the nonzero structure of the
10496: original matrix has not changed from that last call to `MatCreateRedundantMatrix()`.
10498: This routine creates the duplicated matrices in the subcommunicators; you should NOT create them before
10499: calling it.
10501: `PetscSubcommCreate()` can be used to manage the creation of the subcomm but need not be.
10503: .seealso: [](ch_matrices), `Mat`, `MatDestroy()`, `PetscSubcommCreate()`, `PetscSubcomm`
10504: @*/
10505: PetscErrorCode MatCreateRedundantMatrix(Mat mat, PetscInt nsubcomm, MPI_Comm subcomm, MatReuse reuse, Mat *matredundant)
10506: {
10507: MPI_Comm comm;
10508: PetscMPIInt size;
10509: PetscInt mloc_sub, nloc_sub, rstart, rend, M = mat->rmap->N, N = mat->cmap->N, bs = mat->rmap->bs;
10510: Mat_Redundant *redund = NULL;
10511: PetscSubcomm psubcomm = NULL;
10512: MPI_Comm subcomm_in = subcomm;
10513: Mat *matseq;
10514: IS isrow, iscol;
10515: PetscBool newsubcomm = PETSC_FALSE;
10517: PetscFunctionBegin;
10519: if (nsubcomm && reuse == MAT_REUSE_MATRIX) {
10520: PetscAssertPointer(*matredundant, 5);
10522: }
10524: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
10525: if (size == 1 || nsubcomm == 1) {
10526: if (reuse == MAT_INITIAL_MATRIX) {
10527: PetscCall(MatDuplicate(mat, MAT_COPY_VALUES, matredundant));
10528: } else {
10529: 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");
10530: PetscCall(MatCopy(mat, *matredundant, SAME_NONZERO_PATTERN));
10531: }
10532: PetscFunctionReturn(PETSC_SUCCESS);
10533: }
10535: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
10536: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
10537: MatCheckPreallocated(mat, 1);
10539: PetscCall(PetscLogEventBegin(MAT_RedundantMat, mat, 0, 0, 0));
10540: if (subcomm_in == MPI_COMM_NULL && reuse == MAT_INITIAL_MATRIX) { /* get subcomm if user does not provide subcomm */
10541: /* create psubcomm, then get subcomm */
10542: PetscCall(PetscObjectGetComm((PetscObject)mat, &comm));
10543: PetscCallMPI(MPI_Comm_size(comm, &size));
10544: PetscCheck(nsubcomm >= 1 && nsubcomm <= size, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "nsubcomm must between 1 and %d", size);
10546: PetscCall(PetscSubcommCreate(comm, &psubcomm));
10547: PetscCall(PetscSubcommSetNumber(psubcomm, nsubcomm));
10548: PetscCall(PetscSubcommSetType(psubcomm, PETSC_SUBCOMM_CONTIGUOUS));
10549: PetscCall(PetscSubcommSetFromOptions(psubcomm));
10550: PetscCall(PetscCommDuplicate(PetscSubcommChild(psubcomm), &subcomm, NULL));
10551: newsubcomm = PETSC_TRUE;
10552: PetscCall(PetscSubcommDestroy(&psubcomm));
10553: }
10555: /* get isrow, iscol and a local sequential matrix matseq[0] */
10556: if (reuse == MAT_INITIAL_MATRIX) {
10557: mloc_sub = PETSC_DECIDE;
10558: nloc_sub = PETSC_DECIDE;
10559: if (bs < 1) {
10560: PetscCall(PetscSplitOwnership(subcomm, &mloc_sub, &M));
10561: PetscCall(PetscSplitOwnership(subcomm, &nloc_sub, &N));
10562: } else {
10563: PetscCall(PetscSplitOwnershipBlock(subcomm, bs, &mloc_sub, &M));
10564: PetscCall(PetscSplitOwnershipBlock(subcomm, bs, &nloc_sub, &N));
10565: }
10566: PetscCallMPI(MPI_Scan(&mloc_sub, &rend, 1, MPIU_INT, MPI_SUM, subcomm));
10567: rstart = rend - mloc_sub;
10568: PetscCall(ISCreateStride(PETSC_COMM_SELF, mloc_sub, rstart, 1, &isrow));
10569: PetscCall(ISCreateStride(PETSC_COMM_SELF, N, 0, 1, &iscol));
10570: PetscCall(ISSetIdentity(iscol));
10571: } else { /* reuse == MAT_REUSE_MATRIX */
10572: 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");
10573: /* retrieve subcomm */
10574: PetscCall(PetscObjectGetComm((PetscObject)*matredundant, &subcomm));
10575: redund = (*matredundant)->redundant;
10576: isrow = redund->isrow;
10577: iscol = redund->iscol;
10578: matseq = redund->matseq;
10579: }
10580: PetscCall(MatCreateSubMatrices(mat, 1, &isrow, &iscol, reuse, &matseq));
10582: /* get matredundant over subcomm */
10583: if (reuse == MAT_INITIAL_MATRIX) {
10584: PetscCall(MatCreateMPIMatConcatenateSeqMat(subcomm, matseq[0], nloc_sub, reuse, matredundant));
10586: /* create a supporting struct and attach it to C for reuse */
10587: PetscCall(PetscNew(&redund));
10588: (*matredundant)->redundant = redund;
10589: redund->isrow = isrow;
10590: redund->iscol = iscol;
10591: redund->matseq = matseq;
10592: if (newsubcomm) {
10593: redund->subcomm = subcomm;
10594: } else {
10595: redund->subcomm = MPI_COMM_NULL;
10596: }
10597: } else {
10598: PetscCall(MatCreateMPIMatConcatenateSeqMat(subcomm, matseq[0], PETSC_DECIDE, reuse, matredundant));
10599: }
10600: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_HIP)
10601: if (matseq[0]->boundtocpu && matseq[0]->bindingpropagates) {
10602: PetscCall(MatBindToCPU(*matredundant, PETSC_TRUE));
10603: PetscCall(MatSetBindingPropagates(*matredundant, PETSC_TRUE));
10604: }
10605: #endif
10606: PetscCall(PetscLogEventEnd(MAT_RedundantMat, mat, 0, 0, 0));
10607: PetscFunctionReturn(PETSC_SUCCESS);
10608: }
10610: /*@C
10611: MatGetMultiProcBlock - Create multiple 'parallel submatrices' from
10612: a given `Mat`. Each submatrix can span multiple procs.
10614: Collective
10616: Input Parameters:
10617: + mat - the matrix
10618: . subComm - the sub communicator obtained as if by `MPI_Comm_split(PetscObjectComm((PetscObject)mat))`
10619: - scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10621: Output Parameter:
10622: . subMat - parallel sub-matrices each spanning a given `subcomm`
10624: Level: advanced
10626: Notes:
10627: The submatrix partition across processors is dictated by `subComm` a
10628: communicator obtained by `MPI_comm_split()` or via `PetscSubcommCreate()`. The `subComm`
10629: is not restricted to be grouped with consecutive original MPI processes.
10631: Due the `MPI_Comm_split()` usage, the parallel layout of the submatrices
10632: map directly to the layout of the original matrix [wrt the local
10633: row,col partitioning]. So the original 'DiagonalMat' naturally maps
10634: into the 'DiagonalMat' of the `subMat`, hence it is used directly from
10635: the `subMat`. However the offDiagMat looses some columns - and this is
10636: reconstructed with `MatSetValues()`
10638: This is used by `PCBJACOBI` when a single block spans multiple MPI processes.
10640: .seealso: [](ch_matrices), `Mat`, `MatCreateRedundantMatrix()`, `MatCreateSubMatrices()`, `PCBJACOBI`
10641: @*/
10642: PetscErrorCode MatGetMultiProcBlock(Mat mat, MPI_Comm subComm, MatReuse scall, Mat *subMat)
10643: {
10644: PetscMPIInt commsize, subCommSize;
10646: PetscFunctionBegin;
10647: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &commsize));
10648: PetscCallMPI(MPI_Comm_size(subComm, &subCommSize));
10649: PetscCheck(subCommSize <= commsize, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "CommSize %d < SubCommZize %d", commsize, subCommSize);
10651: 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");
10652: PetscCall(PetscLogEventBegin(MAT_GetMultiProcBlock, mat, 0, 0, 0));
10653: PetscUseTypeMethod(mat, getmultiprocblock, subComm, scall, subMat);
10654: PetscCall(PetscLogEventEnd(MAT_GetMultiProcBlock, mat, 0, 0, 0));
10655: PetscFunctionReturn(PETSC_SUCCESS);
10656: }
10658: /*@
10659: MatGetLocalSubMatrix - Gets a reference to a submatrix specified in local numbering
10661: Not Collective
10663: Input Parameters:
10664: + mat - matrix to extract local submatrix from
10665: . isrow - local row indices for submatrix
10666: - iscol - local column indices for submatrix
10668: Output Parameter:
10669: . submat - the submatrix
10671: Level: intermediate
10673: Notes:
10674: `submat` should be disposed of with `MatRestoreLocalSubMatrix()`.
10676: Depending on the format of `mat`, the returned `submat` may not implement `MatMult()`. Its communicator may be
10677: the same as `mat`, it may be `PETSC_COMM_SELF`, or some other sub-communictor of `mat`'s.
10679: `submat` always implements `MatSetValuesLocal()`. If `isrow` and `iscol` have the same block size, then
10680: `MatSetValuesBlockedLocal()` will also be implemented.
10682: `mat` must have had a `ISLocalToGlobalMapping` provided to it with `MatSetLocalToGlobalMapping()`.
10683: Matrices obtained with `DMCreateMatrix()` generally already have the local to global mapping provided.
10685: .seealso: [](ch_matrices), `Mat`, `MatRestoreLocalSubMatrix()`, `MatCreateLocalRef()`, `MatSetLocalToGlobalMapping()`
10686: @*/
10687: PetscErrorCode MatGetLocalSubMatrix(Mat mat, IS isrow, IS iscol, Mat *submat)
10688: {
10689: PetscFunctionBegin;
10693: PetscCheckSameComm(isrow, 2, iscol, 3);
10694: PetscAssertPointer(submat, 4);
10695: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Matrix must have local to global mapping provided before this call");
10697: if (mat->ops->getlocalsubmatrix) {
10698: PetscUseTypeMethod(mat, getlocalsubmatrix, isrow, iscol, submat);
10699: } else {
10700: PetscCall(MatCreateLocalRef(mat, isrow, iscol, submat));
10701: }
10702: PetscFunctionReturn(PETSC_SUCCESS);
10703: }
10705: /*@
10706: MatRestoreLocalSubMatrix - Restores a reference to a submatrix specified in local numbering obtained with `MatGetLocalSubMatrix()`
10708: Not Collective
10710: Input Parameters:
10711: + mat - matrix to extract local submatrix from
10712: . isrow - local row indices for submatrix
10713: . iscol - local column indices for submatrix
10714: - submat - the submatrix
10716: Level: intermediate
10718: .seealso: [](ch_matrices), `Mat`, `MatGetLocalSubMatrix()`
10719: @*/
10720: PetscErrorCode MatRestoreLocalSubMatrix(Mat mat, IS isrow, IS iscol, Mat *submat)
10721: {
10722: PetscFunctionBegin;
10726: PetscCheckSameComm(isrow, 2, iscol, 3);
10727: PetscAssertPointer(submat, 4);
10730: if (mat->ops->restorelocalsubmatrix) {
10731: PetscUseTypeMethod(mat, restorelocalsubmatrix, isrow, iscol, submat);
10732: } else {
10733: PetscCall(MatDestroy(submat));
10734: }
10735: *submat = NULL;
10736: PetscFunctionReturn(PETSC_SUCCESS);
10737: }
10739: /*@
10740: MatFindZeroDiagonals - Finds all the rows of a matrix that have zero or no diagonal entry in the matrix
10742: Collective
10744: Input Parameter:
10745: . mat - the matrix
10747: Output Parameter:
10748: . is - if any rows have zero diagonals this contains the list of them
10750: Level: developer
10752: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
10753: @*/
10754: PetscErrorCode MatFindZeroDiagonals(Mat mat, IS *is)
10755: {
10756: PetscFunctionBegin;
10759: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
10760: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
10762: if (!mat->ops->findzerodiagonals) {
10763: Vec diag;
10764: const PetscScalar *a;
10765: PetscInt *rows;
10766: PetscInt rStart, rEnd, r, nrow = 0;
10768: PetscCall(MatCreateVecs(mat, &diag, NULL));
10769: PetscCall(MatGetDiagonal(mat, diag));
10770: PetscCall(MatGetOwnershipRange(mat, &rStart, &rEnd));
10771: PetscCall(VecGetArrayRead(diag, &a));
10772: for (r = 0; r < rEnd - rStart; ++r)
10773: if (a[r] == 0.0) ++nrow;
10774: PetscCall(PetscMalloc1(nrow, &rows));
10775: nrow = 0;
10776: for (r = 0; r < rEnd - rStart; ++r)
10777: if (a[r] == 0.0) rows[nrow++] = r + rStart;
10778: PetscCall(VecRestoreArrayRead(diag, &a));
10779: PetscCall(VecDestroy(&diag));
10780: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)mat), nrow, rows, PETSC_OWN_POINTER, is));
10781: } else {
10782: PetscUseTypeMethod(mat, findzerodiagonals, is);
10783: }
10784: PetscFunctionReturn(PETSC_SUCCESS);
10785: }
10787: /*@
10788: MatFindOffBlockDiagonalEntries - Finds all the rows of a matrix that have entries outside of the main diagonal block (defined by the matrix block size)
10790: Collective
10792: Input Parameter:
10793: . mat - the matrix
10795: Output Parameter:
10796: . is - contains the list of rows with off block diagonal entries
10798: Level: developer
10800: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
10801: @*/
10802: PetscErrorCode MatFindOffBlockDiagonalEntries(Mat mat, IS *is)
10803: {
10804: PetscFunctionBegin;
10807: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
10808: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
10810: PetscUseTypeMethod(mat, findoffblockdiagonalentries, is);
10811: PetscFunctionReturn(PETSC_SUCCESS);
10812: }
10814: /*@C
10815: MatInvertBlockDiagonal - Inverts the block diagonal entries.
10817: Collective; No Fortran Support
10819: Input Parameter:
10820: . mat - the matrix
10822: Output Parameter:
10823: . values - the block inverses in column major order (FORTRAN-like)
10825: Level: advanced
10827: Notes:
10828: The size of the blocks is determined by the block size of the matrix.
10830: The blocks never overlap between two MPI processes, use `MatInvertVariableBlockEnvelope()` for that case
10832: The blocks all have the same size, use `MatInvertVariableBlockDiagonal()` for variable block size
10834: .seealso: [](ch_matrices), `Mat`, `MatInvertVariableBlockEnvelope()`, `MatInvertBlockDiagonalMat()`
10835: @*/
10836: PetscErrorCode MatInvertBlockDiagonal(Mat mat, const PetscScalar *values[])
10837: {
10838: PetscFunctionBegin;
10840: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
10841: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
10842: PetscUseTypeMethod(mat, invertblockdiagonal, values);
10843: PetscFunctionReturn(PETSC_SUCCESS);
10844: }
10846: /*@
10847: MatInvertVariableBlockDiagonal - Inverts the point block diagonal entries.
10849: Collective; No Fortran Support
10851: Input Parameters:
10852: + mat - the matrix
10853: . nblocks - the number of blocks on the process, set with `MatSetVariableBlockSizes()`
10854: - bsizes - the size of each block on the process, set with `MatSetVariableBlockSizes()`
10856: Output Parameter:
10857: . values - the block inverses in column major order (FORTRAN-like)
10859: Level: advanced
10861: Notes:
10862: Use `MatInvertBlockDiagonal()` if all blocks have the same size
10864: The blocks never overlap between two MPI processes, use `MatInvertVariableBlockEnvelope()` for that case
10866: .seealso: [](ch_matrices), `Mat`, `MatInvertBlockDiagonal()`, `MatSetVariableBlockSizes()`, `MatInvertVariableBlockEnvelope()`
10867: @*/
10868: PetscErrorCode MatInvertVariableBlockDiagonal(Mat mat, PetscInt nblocks, const PetscInt bsizes[], PetscScalar values[])
10869: {
10870: PetscFunctionBegin;
10872: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
10873: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
10874: PetscUseTypeMethod(mat, invertvariableblockdiagonal, nblocks, bsizes, values);
10875: PetscFunctionReturn(PETSC_SUCCESS);
10876: }
10878: /*@
10879: MatInvertBlockDiagonalMat - set the values of matrix C to be the inverted block diagonal of matrix A
10881: Collective
10883: Input Parameters:
10884: + A - the matrix
10885: - C - matrix with inverted block diagonal of `A`. This matrix should be created and may have its type set.
10887: Level: advanced
10889: Note:
10890: The blocksize of the matrix is used to determine the blocks on the diagonal of `C`
10892: .seealso: [](ch_matrices), `Mat`, `MatInvertBlockDiagonal()`
10893: @*/
10894: PetscErrorCode MatInvertBlockDiagonalMat(Mat A, Mat C)
10895: {
10896: const PetscScalar *vals;
10897: PetscInt *dnnz;
10898: PetscInt m, rstart, rend, bs, i, j;
10900: PetscFunctionBegin;
10901: PetscCall(MatInvertBlockDiagonal(A, &vals));
10902: PetscCall(MatGetBlockSize(A, &bs));
10903: PetscCall(MatGetLocalSize(A, &m, NULL));
10904: PetscCall(MatSetLayouts(C, A->rmap, A->cmap));
10905: PetscCall(PetscMalloc1(m / bs, &dnnz));
10906: for (j = 0; j < m / bs; j++) dnnz[j] = 1;
10907: PetscCall(MatXAIJSetPreallocation(C, bs, dnnz, NULL, NULL, NULL));
10908: PetscCall(PetscFree(dnnz));
10909: PetscCall(MatGetOwnershipRange(C, &rstart, &rend));
10910: PetscCall(MatSetOption(C, MAT_ROW_ORIENTED, PETSC_FALSE));
10911: for (i = rstart / bs; i < rend / bs; i++) PetscCall(MatSetValuesBlocked(C, 1, &i, 1, &i, &vals[(i - rstart / bs) * bs * bs], INSERT_VALUES));
10912: PetscCall(MatAssemblyBegin(C, MAT_FINAL_ASSEMBLY));
10913: PetscCall(MatAssemblyEnd(C, MAT_FINAL_ASSEMBLY));
10914: PetscCall(MatSetOption(C, MAT_ROW_ORIENTED, PETSC_TRUE));
10915: PetscFunctionReturn(PETSC_SUCCESS);
10916: }
10918: /*@
10919: MatTransposeColoringDestroy - Destroys a coloring context for matrix product $C = A*B^T$ that was created
10920: via `MatTransposeColoringCreate()`.
10922: Collective
10924: Input Parameter:
10925: . c - coloring context
10927: Level: intermediate
10929: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringCreate()`
10930: @*/
10931: PetscErrorCode MatTransposeColoringDestroy(MatTransposeColoring *c)
10932: {
10933: MatTransposeColoring matcolor = *c;
10935: PetscFunctionBegin;
10936: if (!matcolor) PetscFunctionReturn(PETSC_SUCCESS);
10937: if (--((PetscObject)matcolor)->refct > 0) {
10938: matcolor = NULL;
10939: PetscFunctionReturn(PETSC_SUCCESS);
10940: }
10942: PetscCall(PetscFree3(matcolor->ncolumns, matcolor->nrows, matcolor->colorforrow));
10943: PetscCall(PetscFree(matcolor->rows));
10944: PetscCall(PetscFree(matcolor->den2sp));
10945: PetscCall(PetscFree(matcolor->colorforcol));
10946: PetscCall(PetscFree(matcolor->columns));
10947: if (matcolor->brows > 0) PetscCall(PetscFree(matcolor->lstart));
10948: PetscCall(PetscHeaderDestroy(c));
10949: PetscFunctionReturn(PETSC_SUCCESS);
10950: }
10952: /*@
10953: MatTransColoringApplySpToDen - Given a symbolic matrix product $C = A*B^T$ for which
10954: a `MatTransposeColoring` context has been created, computes a dense $B^T$ by applying
10955: `MatTransposeColoring` to sparse `B`.
10957: Collective
10959: Input Parameters:
10960: + coloring - coloring context created with `MatTransposeColoringCreate()`
10961: - B - sparse matrix
10963: Output Parameter:
10964: . Btdense - dense matrix $B^T$
10966: Level: developer
10968: Note:
10969: These are used internally for some implementations of `MatRARt()`
10971: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringCreate()`, `MatTransposeColoringDestroy()`, `MatTransColoringApplyDenToSp()`
10972: @*/
10973: PetscErrorCode MatTransColoringApplySpToDen(MatTransposeColoring coloring, Mat B, Mat Btdense)
10974: {
10975: PetscFunctionBegin;
10980: PetscCall((*B->ops->transcoloringapplysptoden)(coloring, B, Btdense));
10981: PetscFunctionReturn(PETSC_SUCCESS);
10982: }
10984: /*@
10985: MatTransColoringApplyDenToSp - Given a symbolic matrix product $C_{sp} = A*B^T$ for which
10986: a `MatTransposeColoring` context has been created and a dense matrix $C_{den} = A*B^T_{dense}$
10987: in which `B^T_{dens}` is obtained from `MatTransColoringApplySpToDen()`, recover sparse matrix
10988: $C_{sp}$ from $C_{den}$.
10990: Collective
10992: Input Parameters:
10993: + matcoloring - coloring context created with `MatTransposeColoringCreate()`
10994: - Cden - matrix product of a sparse matrix and a dense matrix Btdense
10996: Output Parameter:
10997: . Csp - sparse matrix
10999: Level: developer
11001: Note:
11002: These are used internally for some implementations of `MatRARt()`
11004: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringCreate()`, `MatTransposeColoringDestroy()`, `MatTransColoringApplySpToDen()`
11005: @*/
11006: PetscErrorCode MatTransColoringApplyDenToSp(MatTransposeColoring matcoloring, Mat Cden, Mat Csp)
11007: {
11008: PetscFunctionBegin;
11013: PetscCall((*Csp->ops->transcoloringapplydentosp)(matcoloring, Cden, Csp));
11014: PetscCall(MatAssemblyBegin(Csp, MAT_FINAL_ASSEMBLY));
11015: PetscCall(MatAssemblyEnd(Csp, MAT_FINAL_ASSEMBLY));
11016: PetscFunctionReturn(PETSC_SUCCESS);
11017: }
11019: /*@
11020: MatTransposeColoringCreate - Creates a matrix coloring context for the matrix product $C = A*B^T$.
11022: Collective
11024: Input Parameters:
11025: + mat - the matrix product C
11026: - iscoloring - the coloring of the matrix; usually obtained with `MatColoringCreate()` or `DMCreateColoring()`
11028: Output Parameter:
11029: . color - the new coloring context
11031: Level: intermediate
11033: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringDestroy()`, `MatTransColoringApplySpToDen()`,
11034: `MatTransColoringApplyDenToSp()`
11035: @*/
11036: PetscErrorCode MatTransposeColoringCreate(Mat mat, ISColoring iscoloring, MatTransposeColoring *color)
11037: {
11038: MatTransposeColoring c;
11039: MPI_Comm comm;
11041: PetscFunctionBegin;
11042: PetscAssertPointer(color, 3);
11044: PetscCall(PetscLogEventBegin(MAT_TransposeColoringCreate, mat, 0, 0, 0));
11045: PetscCall(PetscObjectGetComm((PetscObject)mat, &comm));
11046: PetscCall(PetscHeaderCreate(c, MAT_TRANSPOSECOLORING_CLASSID, "MatTransposeColoring", "Matrix product C=A*B^T via coloring", "Mat", comm, MatTransposeColoringDestroy, NULL));
11047: c->ctype = iscoloring->ctype;
11048: PetscUseTypeMethod(mat, transposecoloringcreate, iscoloring, c);
11049: *color = c;
11050: PetscCall(PetscLogEventEnd(MAT_TransposeColoringCreate, mat, 0, 0, 0));
11051: PetscFunctionReturn(PETSC_SUCCESS);
11052: }
11054: /*@
11055: MatGetNonzeroState - Returns a 64-bit integer representing the current state of nonzeros in the matrix. If the
11056: matrix has had new nonzero locations added to (or removed from) the matrix since the previous call, the value will be larger.
11058: Not Collective
11060: Input Parameter:
11061: . mat - the matrix
11063: Output Parameter:
11064: . state - the current state
11066: Level: intermediate
11068: Notes:
11069: You can only compare states from two different calls to the SAME matrix, you cannot compare calls between
11070: different matrices
11072: Use `PetscObjectStateGet()` to check for changes to the numerical values in a matrix
11074: Use the result of `PetscObjectGetId()` to compare if a previously checked matrix is the same as the current matrix, do not compare object pointers.
11076: .seealso: [](ch_matrices), `Mat`, `PetscObjectStateGet()`, `PetscObjectGetId()`
11077: @*/
11078: PetscErrorCode MatGetNonzeroState(Mat mat, PetscObjectState *state)
11079: {
11080: PetscFunctionBegin;
11082: *state = mat->nonzerostate;
11083: PetscFunctionReturn(PETSC_SUCCESS);
11084: }
11086: /*@
11087: MatCreateMPIMatConcatenateSeqMat - Creates a single large PETSc matrix by concatenating sequential
11088: matrices from each processor
11090: Collective
11092: Input Parameters:
11093: + comm - the communicators the parallel matrix will live on
11094: . seqmat - the input sequential matrices
11095: . n - number of local columns (or `PETSC_DECIDE`)
11096: - reuse - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
11098: Output Parameter:
11099: . mpimat - the parallel matrix generated
11101: Level: developer
11103: Note:
11104: The number of columns of the matrix in EACH processor MUST be the same.
11106: .seealso: [](ch_matrices), `Mat`
11107: @*/
11108: PetscErrorCode MatCreateMPIMatConcatenateSeqMat(MPI_Comm comm, Mat seqmat, PetscInt n, MatReuse reuse, Mat *mpimat)
11109: {
11110: PetscMPIInt size;
11112: PetscFunctionBegin;
11113: PetscCallMPI(MPI_Comm_size(comm, &size));
11114: if (size == 1) {
11115: if (reuse == MAT_INITIAL_MATRIX) {
11116: PetscCall(MatDuplicate(seqmat, MAT_COPY_VALUES, mpimat));
11117: } else {
11118: PetscCall(MatCopy(seqmat, *mpimat, SAME_NONZERO_PATTERN));
11119: }
11120: PetscFunctionReturn(PETSC_SUCCESS);
11121: }
11123: 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");
11125: PetscCall(PetscLogEventBegin(MAT_Merge, seqmat, 0, 0, 0));
11126: PetscCall((*seqmat->ops->creatempimatconcatenateseqmat)(comm, seqmat, n, reuse, mpimat));
11127: PetscCall(PetscLogEventEnd(MAT_Merge, seqmat, 0, 0, 0));
11128: PetscFunctionReturn(PETSC_SUCCESS);
11129: }
11131: /*@
11132: MatSubdomainsCreateCoalesce - Creates index subdomains by coalescing adjacent MPI processes' ownership ranges.
11134: Collective
11136: Input Parameters:
11137: + A - the matrix to create subdomains from
11138: - N - requested number of subdomains
11140: Output Parameters:
11141: + n - number of subdomains resulting on this MPI process
11142: - iss - `IS` list with indices of subdomains on this MPI process
11144: Level: advanced
11146: Note:
11147: The number of subdomains must be smaller than the communicator size
11149: .seealso: [](ch_matrices), `Mat`, `IS`
11150: @*/
11151: PetscErrorCode MatSubdomainsCreateCoalesce(Mat A, PetscInt N, PetscInt *n, IS *iss[])
11152: {
11153: MPI_Comm comm, subcomm;
11154: PetscMPIInt size, rank, color;
11155: PetscInt rstart, rend, k;
11157: PetscFunctionBegin;
11158: PetscCall(PetscObjectGetComm((PetscObject)A, &comm));
11159: PetscCallMPI(MPI_Comm_size(comm, &size));
11160: PetscCallMPI(MPI_Comm_rank(comm, &rank));
11161: PetscCheck(N >= 1 && N < (PetscInt)size, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "number of subdomains must be > 0 and < %d, got N = %" PetscInt_FMT, size, N);
11162: *n = 1;
11163: k = ((PetscInt)size) / N + ((PetscInt)size % N > 0); /* There are up to k ranks to a color */
11164: color = rank / k;
11165: PetscCallMPI(MPI_Comm_split(comm, color, rank, &subcomm));
11166: PetscCall(PetscMalloc1(1, iss));
11167: PetscCall(MatGetOwnershipRange(A, &rstart, &rend));
11168: PetscCall(ISCreateStride(subcomm, rend - rstart, rstart, 1, iss[0]));
11169: PetscCallMPI(MPI_Comm_free(&subcomm));
11170: PetscFunctionReturn(PETSC_SUCCESS);
11171: }
11173: /*@
11174: MatGalerkin - Constructs the coarse grid problem matrix via Galerkin projection.
11176: If the interpolation and restriction operators are the same, uses `MatPtAP()`.
11177: If they are not the same, uses `MatMatMatMult()`.
11179: Once the coarse grid problem is constructed, correct for interpolation operators
11180: that are not of full rank, which can legitimately happen in the case of non-nested
11181: geometric multigrid.
11183: Input Parameters:
11184: + restrct - restriction operator
11185: . dA - fine grid matrix
11186: . interpolate - interpolation operator
11187: . reuse - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
11188: - fill - expected fill, use `PETSC_DETERMINE` or `PETSC_DETERMINE` if you do not have a good estimate
11190: Output Parameter:
11191: . A - the Galerkin coarse matrix
11193: Options Database Key:
11194: . -pc_mg_galerkin <both,pmat,mat,none> - for what matrices the Galerkin process should be used
11196: Level: developer
11198: Note:
11199: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
11201: .seealso: [](ch_matrices), `Mat`, `MatPtAP()`, `MatMatMatMult()`
11202: @*/
11203: PetscErrorCode MatGalerkin(Mat restrct, Mat dA, Mat interpolate, MatReuse reuse, PetscReal fill, Mat *A)
11204: {
11205: IS zerorows;
11206: Vec diag;
11208: PetscFunctionBegin;
11209: PetscCheck(reuse != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
11210: /* Construct the coarse grid matrix */
11211: if (interpolate == restrct) {
11212: PetscCall(MatPtAP(dA, interpolate, reuse, fill, A));
11213: } else {
11214: PetscCall(MatMatMatMult(restrct, dA, interpolate, reuse, fill, A));
11215: }
11217: /* If the interpolation matrix is not of full rank, A will have zero rows.
11218: This can legitimately happen in the case of non-nested geometric multigrid.
11219: In that event, we set the rows of the matrix to the rows of the identity,
11220: ignoring the equations (as the RHS will also be zero). */
11222: PetscCall(MatFindZeroRows(*A, &zerorows));
11224: if (zerorows != NULL) { /* if there are any zero rows */
11225: PetscCall(MatCreateVecs(*A, &diag, NULL));
11226: PetscCall(MatGetDiagonal(*A, diag));
11227: PetscCall(VecISSet(diag, zerorows, 1.0));
11228: PetscCall(MatDiagonalSet(*A, diag, INSERT_VALUES));
11229: PetscCall(VecDestroy(&diag));
11230: PetscCall(ISDestroy(&zerorows));
11231: }
11232: PetscFunctionReturn(PETSC_SUCCESS);
11233: }
11235: /*@C
11236: MatSetOperation - Allows user to set a matrix operation for any matrix type
11238: Logically Collective
11240: Input Parameters:
11241: + mat - the matrix
11242: . op - the name of the operation
11243: - f - the function that provides the operation
11245: Level: developer
11247: Example Usage:
11248: .vb
11249: extern PetscErrorCode usermult(Mat, Vec, Vec);
11251: PetscCall(MatCreateXXX(comm, ..., &A));
11252: PetscCall(MatSetOperation(A, MATOP_MULT, (PetscVoidFn *)usermult));
11253: .ve
11255: Notes:
11256: See the file `include/petscmat.h` for a complete list of matrix
11257: operations, which all have the form MATOP_<OPERATION>, where
11258: <OPERATION> is the name (in all capital letters) of the
11259: user interface routine (e.g., `MatMult()` -> `MATOP_MULT`).
11261: All user-provided functions (except for `MATOP_DESTROY`) should have the same calling
11262: sequence as the usual matrix interface routines, since they
11263: are intended to be accessed via the usual matrix interface
11264: routines, e.g.,
11265: .vb
11266: MatMult(Mat, Vec, Vec) -> usermult(Mat, Vec, Vec)
11267: .ve
11269: In particular each function MUST return `PETSC_SUCCESS` on success and
11270: nonzero on failure.
11272: This routine is distinct from `MatShellSetOperation()` in that it can be called on any matrix type.
11274: .seealso: [](ch_matrices), `Mat`, `MatGetOperation()`, `MatCreateShell()`, `MatShellSetContext()`, `MatShellSetOperation()`
11275: @*/
11276: PetscErrorCode MatSetOperation(Mat mat, MatOperation op, void (*f)(void))
11277: {
11278: PetscFunctionBegin;
11280: if (op == MATOP_VIEW && !mat->ops->viewnative && f != (void (*)(void))mat->ops->view) mat->ops->viewnative = mat->ops->view;
11281: (((void (**)(void))mat->ops)[op]) = f;
11282: PetscFunctionReturn(PETSC_SUCCESS);
11283: }
11285: /*@C
11286: MatGetOperation - Gets a matrix operation for any matrix type.
11288: Not Collective
11290: Input Parameters:
11291: + mat - the matrix
11292: - op - the name of the operation
11294: Output Parameter:
11295: . f - the function that provides the operation
11297: Level: developer
11299: Example Usage:
11300: .vb
11301: PetscErrorCode (*usermult)(Mat, Vec, Vec);
11303: MatGetOperation(A, MATOP_MULT, (void (**)(void))&usermult);
11304: .ve
11306: Notes:
11307: See the file include/petscmat.h for a complete list of matrix
11308: operations, which all have the form MATOP_<OPERATION>, where
11309: <OPERATION> is the name (in all capital letters) of the
11310: user interface routine (e.g., `MatMult()` -> `MATOP_MULT`).
11312: This routine is distinct from `MatShellGetOperation()` in that it can be called on any matrix type.
11314: .seealso: [](ch_matrices), `Mat`, `MatSetOperation()`, `MatCreateShell()`, `MatShellGetContext()`, `MatShellGetOperation()`
11315: @*/
11316: PetscErrorCode MatGetOperation(Mat mat, MatOperation op, void (**f)(void))
11317: {
11318: PetscFunctionBegin;
11320: *f = (((void (**)(void))mat->ops)[op]);
11321: PetscFunctionReturn(PETSC_SUCCESS);
11322: }
11324: /*@
11325: MatHasOperation - Determines whether the given matrix supports the particular operation.
11327: Not Collective
11329: Input Parameters:
11330: + mat - the matrix
11331: - op - the operation, for example, `MATOP_GET_DIAGONAL`
11333: Output Parameter:
11334: . has - either `PETSC_TRUE` or `PETSC_FALSE`
11336: Level: advanced
11338: Note:
11339: See `MatSetOperation()` for additional discussion on naming convention and usage of `op`.
11341: .seealso: [](ch_matrices), `Mat`, `MatCreateShell()`, `MatGetOperation()`, `MatSetOperation()`
11342: @*/
11343: PetscErrorCode MatHasOperation(Mat mat, MatOperation op, PetscBool *has)
11344: {
11345: PetscFunctionBegin;
11347: PetscAssertPointer(has, 3);
11348: if (mat->ops->hasoperation) {
11349: PetscUseTypeMethod(mat, hasoperation, op, has);
11350: } else {
11351: if (((void **)mat->ops)[op]) *has = PETSC_TRUE;
11352: else {
11353: *has = PETSC_FALSE;
11354: if (op == MATOP_CREATE_SUBMATRIX) {
11355: PetscMPIInt size;
11357: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
11358: if (size == 1) PetscCall(MatHasOperation(mat, MATOP_CREATE_SUBMATRICES, has));
11359: }
11360: }
11361: }
11362: PetscFunctionReturn(PETSC_SUCCESS);
11363: }
11365: /*@
11366: MatHasCongruentLayouts - Determines whether the rows and columns layouts of the matrix are congruent
11368: Collective
11370: Input Parameter:
11371: . mat - the matrix
11373: Output Parameter:
11374: . cong - either `PETSC_TRUE` or `PETSC_FALSE`
11376: Level: beginner
11378: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatSetSizes()`, `PetscLayout`
11379: @*/
11380: PetscErrorCode MatHasCongruentLayouts(Mat mat, PetscBool *cong)
11381: {
11382: PetscFunctionBegin;
11385: PetscAssertPointer(cong, 2);
11386: if (!mat->rmap || !mat->cmap) {
11387: *cong = mat->rmap == mat->cmap ? PETSC_TRUE : PETSC_FALSE;
11388: PetscFunctionReturn(PETSC_SUCCESS);
11389: }
11390: if (mat->congruentlayouts == PETSC_DECIDE) { /* first time we compare rows and cols layouts */
11391: PetscCall(PetscLayoutSetUp(mat->rmap));
11392: PetscCall(PetscLayoutSetUp(mat->cmap));
11393: PetscCall(PetscLayoutCompare(mat->rmap, mat->cmap, cong));
11394: if (*cong) mat->congruentlayouts = 1;
11395: else mat->congruentlayouts = 0;
11396: } else *cong = mat->congruentlayouts ? PETSC_TRUE : PETSC_FALSE;
11397: PetscFunctionReturn(PETSC_SUCCESS);
11398: }
11400: PetscErrorCode MatSetInf(Mat A)
11401: {
11402: PetscFunctionBegin;
11403: PetscUseTypeMethod(A, setinf);
11404: PetscFunctionReturn(PETSC_SUCCESS);
11405: }
11407: /*@
11408: 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
11409: and possibly removes small values from the graph structure.
11411: Collective
11413: Input Parameters:
11414: + A - the matrix
11415: . sym - `PETSC_TRUE` indicates that the graph should be symmetrized
11416: . scale - `PETSC_TRUE` indicates that the graph edge weights should be symmetrically scaled with the diagonal entry
11417: . filter - filter value - < 0: does nothing; == 0: removes only 0.0 entries; otherwise: removes entries with abs(entries) <= value
11418: . num_idx - size of 'index' array
11419: - index - array of block indices to use for graph strength of connection weight
11421: Output Parameter:
11422: . graph - the resulting graph
11424: Level: advanced
11426: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `PCGAMG`
11427: @*/
11428: PetscErrorCode MatCreateGraph(Mat A, PetscBool sym, PetscBool scale, PetscReal filter, PetscInt num_idx, PetscInt index[], Mat *graph)
11429: {
11430: PetscFunctionBegin;
11434: PetscAssertPointer(graph, 7);
11435: PetscCall(PetscLogEventBegin(MAT_CreateGraph, A, 0, 0, 0));
11436: PetscUseTypeMethod(A, creategraph, sym, scale, filter, num_idx, index, graph);
11437: PetscCall(PetscLogEventEnd(MAT_CreateGraph, A, 0, 0, 0));
11438: PetscFunctionReturn(PETSC_SUCCESS);
11439: }
11441: /*@
11442: MatEliminateZeros - eliminate the nondiagonal zero entries in place from the nonzero structure of a sparse `Mat` in place,
11443: meaning the same memory is used for the matrix, and no new memory is allocated.
11445: Collective
11447: Input Parameters:
11448: + A - the matrix
11449: - 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
11451: Level: intermediate
11453: Developer Note:
11454: The entries in the sparse matrix data structure are shifted to fill in the unneeded locations in the data. Thus the end
11455: of the arrays in the data structure are unneeded.
11457: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatCreateGraph()`, `MatFilter()`
11458: @*/
11459: PetscErrorCode MatEliminateZeros(Mat A, PetscBool keep)
11460: {
11461: PetscFunctionBegin;
11463: PetscUseTypeMethod(A, eliminatezeros, keep);
11464: PetscFunctionReturn(PETSC_SUCCESS);
11465: }