Actual source code: umfpack.c
2: /*
3: Provides an interface to the UMFPACK sparse solver available through SuiteSparse version 4.2.1
5: When build with PETSC_USE_64BIT_INDICES this will use Suitesparse_long as the
6: integer type in UMFPACK, otherwise it will use int. This means
7: all integers in this file as simply declared as PetscInt. Also it means
8: that one cannot use 64BIT_INDICES on 32-bit pointer systems [as Suitesparse_long is 32-bit only]
10: */
11: #include <../src/mat/impls/aij/seq/aij.h>
13: #if defined(PETSC_USE_64BIT_INDICES)
14: #if defined(PETSC_USE_COMPLEX)
15: #define umfpack_UMF_free_symbolic umfpack_zl_free_symbolic
16: #define umfpack_UMF_free_numeric umfpack_zl_free_numeric
17: /* the type casts are needed because PetscInt is long long while SuiteSparse_long is long and compilers warn even when they are identical */
18: #define umfpack_UMF_wsolve(a, b, c, d, e, f, g, h, i, j, k, l, m, n) umfpack_zl_wsolve(a, (SuiteSparse_long *)b, (SuiteSparse_long *)c, d, e, f, g, h, i, (SuiteSparse_long *)j, k, l, (SuiteSparse_long *)m, n)
19: #define umfpack_UMF_numeric(a, b, c, d, e, f, g, h) umfpack_zl_numeric((SuiteSparse_long *)a, (SuiteSparse_long *)b, c, d, e, f, g, h)
20: #define umfpack_UMF_report_numeric umfpack_zl_report_numeric
21: #define umfpack_UMF_report_control umfpack_zl_report_control
22: #define umfpack_UMF_report_status umfpack_zl_report_status
23: #define umfpack_UMF_report_info umfpack_zl_report_info
24: #define umfpack_UMF_report_symbolic umfpack_zl_report_symbolic
25: #define umfpack_UMF_qsymbolic(a, b, c, d, e, f, g, h, i, j) umfpack_zl_qsymbolic(a, b, (SuiteSparse_long *)c, (SuiteSparse_long *)d, e, f, (SuiteSparse_long *)g, h, i, j)
26: #define umfpack_UMF_symbolic(a, b, c, d, e, f, g, h, i) umfpack_zl_symbolic(a, b, (SuiteSparse_long *)c, (SuiteSparse_long *)d, e, f, g, h, i)
27: #define umfpack_UMF_defaults umfpack_zl_defaults
29: #else
30: #define umfpack_UMF_free_symbolic umfpack_dl_free_symbolic
31: #define umfpack_UMF_free_numeric umfpack_dl_free_numeric
32: #define umfpack_UMF_wsolve(a, b, c, d, e, f, g, h, i, j, k) umfpack_dl_wsolve(a, (SuiteSparse_long *)b, (SuiteSparse_long *)c, d, e, f, g, h, i, (SuiteSparse_long *)j, k)
33: #define umfpack_UMF_numeric(a, b, c, d, e, f, g) umfpack_dl_numeric((SuiteSparse_long *)a, (SuiteSparse_long *)b, c, d, e, f, g)
34: #define umfpack_UMF_report_numeric umfpack_dl_report_numeric
35: #define umfpack_UMF_report_control umfpack_dl_report_control
36: #define umfpack_UMF_report_status umfpack_dl_report_status
37: #define umfpack_UMF_report_info umfpack_dl_report_info
38: #define umfpack_UMF_report_symbolic umfpack_dl_report_symbolic
39: #define umfpack_UMF_qsymbolic(a, b, c, d, e, f, g, h, i) umfpack_dl_qsymbolic(a, b, (SuiteSparse_long *)c, (SuiteSparse_long *)d, e, (SuiteSparse_long *)f, g, h, i)
40: #define umfpack_UMF_symbolic(a, b, c, d, e, f, g, h) umfpack_dl_symbolic(a, b, (SuiteSparse_long *)c, (SuiteSparse_long *)d, e, f, g, h)
41: #define umfpack_UMF_defaults umfpack_dl_defaults
42: #endif
44: #else
45: #if defined(PETSC_USE_COMPLEX)
46: #define umfpack_UMF_free_symbolic umfpack_zi_free_symbolic
47: #define umfpack_UMF_free_numeric umfpack_zi_free_numeric
48: #define umfpack_UMF_wsolve umfpack_zi_wsolve
49: #define umfpack_UMF_numeric umfpack_zi_numeric
50: #define umfpack_UMF_report_numeric umfpack_zi_report_numeric
51: #define umfpack_UMF_report_control umfpack_zi_report_control
52: #define umfpack_UMF_report_status umfpack_zi_report_status
53: #define umfpack_UMF_report_info umfpack_zi_report_info
54: #define umfpack_UMF_report_symbolic umfpack_zi_report_symbolic
55: #define umfpack_UMF_qsymbolic umfpack_zi_qsymbolic
56: #define umfpack_UMF_symbolic umfpack_zi_symbolic
57: #define umfpack_UMF_defaults umfpack_zi_defaults
59: #else
60: #define umfpack_UMF_free_symbolic umfpack_di_free_symbolic
61: #define umfpack_UMF_free_numeric umfpack_di_free_numeric
62: #define umfpack_UMF_wsolve umfpack_di_wsolve
63: #define umfpack_UMF_numeric umfpack_di_numeric
64: #define umfpack_UMF_report_numeric umfpack_di_report_numeric
65: #define umfpack_UMF_report_control umfpack_di_report_control
66: #define umfpack_UMF_report_status umfpack_di_report_status
67: #define umfpack_UMF_report_info umfpack_di_report_info
68: #define umfpack_UMF_report_symbolic umfpack_di_report_symbolic
69: #define umfpack_UMF_qsymbolic umfpack_di_qsymbolic
70: #define umfpack_UMF_symbolic umfpack_di_symbolic
71: #define umfpack_UMF_defaults umfpack_di_defaults
72: #endif
73: #endif
75: EXTERN_C_BEGIN
76: #include <umfpack.h>
77: EXTERN_C_END
79: static const char *const UmfpackOrderingTypes[] = {"CHOLMOD", "AMD", "GIVEN", "METIS", "BEST", "NONE", "USER", "UmfpackOrderingTypes", "UMFPACK_ORDERING_", 0};
81: typedef struct {
82: void *Symbolic, *Numeric;
83: double Info[UMFPACK_INFO], Control[UMFPACK_CONTROL], *W;
84: PetscInt *Wi, *perm_c;
85: Mat A; /* Matrix used for factorization */
86: MatStructure flg;
88: /* Flag to clean up UMFPACK objects during Destroy */
89: PetscBool CleanUpUMFPACK;
90: } Mat_UMFPACK;
92: static PetscErrorCode MatDestroy_UMFPACK(Mat A)
93: {
94: Mat_UMFPACK *lu = (Mat_UMFPACK *)A->data;
96: PetscFunctionBegin;
97: if (lu->CleanUpUMFPACK) {
98: umfpack_UMF_free_symbolic(&lu->Symbolic);
99: umfpack_UMF_free_numeric(&lu->Numeric);
100: PetscCall(PetscFree(lu->Wi));
101: PetscCall(PetscFree(lu->W));
102: PetscCall(PetscFree(lu->perm_c));
103: }
104: PetscCall(MatDestroy(&lu->A));
105: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL));
106: PetscCall(PetscFree(A->data));
107: PetscFunctionReturn(PETSC_SUCCESS);
108: }
110: static PetscErrorCode MatSolve_UMFPACK_Private(Mat A, Vec b, Vec x, int uflag)
111: {
112: Mat_UMFPACK *lu = (Mat_UMFPACK *)A->data;
113: Mat_SeqAIJ *a = (Mat_SeqAIJ *)lu->A->data;
114: PetscScalar *av = a->a, *xa;
115: const PetscScalar *ba;
116: PetscInt *ai = a->i, *aj = a->j, status;
117: static PetscBool cite = PETSC_FALSE;
119: PetscFunctionBegin;
120: if (!A->rmap->n) PetscFunctionReturn(PETSC_SUCCESS);
121: PetscCall(PetscCitationsRegister("@article{davis2004algorithm,\n title={Algorithm 832: {UMFPACK} V4.3---An Unsymmetric-Pattern Multifrontal Method},\n author={Davis, Timothy A},\n journal={ACM Transactions on Mathematical Software (TOMS)},\n "
122: "volume={30},\n number={2},\n pages={196--199},\n year={2004},\n publisher={ACM}\n}\n",
123: &cite));
124: /* solve Ax = b by umfpack_*_wsolve */
126: if (!lu->Wi) { /* first time, allocate working space for wsolve */
127: PetscCall(PetscMalloc1(A->rmap->n, &lu->Wi));
128: PetscCall(PetscMalloc1(5 * A->rmap->n, &lu->W));
129: }
131: PetscCall(VecGetArrayRead(b, &ba));
132: PetscCall(VecGetArray(x, &xa));
133: #if defined(PETSC_USE_COMPLEX)
134: status = umfpack_UMF_wsolve(uflag, ai, aj, (PetscReal *)av, NULL, (PetscReal *)xa, NULL, (PetscReal *)ba, NULL, lu->Numeric, lu->Control, lu->Info, lu->Wi, lu->W);
135: #else
136: status = umfpack_UMF_wsolve(uflag, ai, aj, av, xa, ba, lu->Numeric, lu->Control, lu->Info, lu->Wi, lu->W);
137: #endif
138: umfpack_UMF_report_info(lu->Control, lu->Info);
139: if (status < 0) {
140: umfpack_UMF_report_status(lu->Control, status);
141: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "umfpack_UMF_wsolve failed");
142: }
144: PetscCall(VecRestoreArrayRead(b, &ba));
145: PetscCall(VecRestoreArray(x, &xa));
146: PetscFunctionReturn(PETSC_SUCCESS);
147: }
149: static PetscErrorCode MatSolve_UMFPACK(Mat A, Vec b, Vec x)
150: {
151: PetscFunctionBegin;
152: /* We gave UMFPACK the algebraic transpose (because it assumes column alignment) */
153: PetscCall(MatSolve_UMFPACK_Private(A, b, x, UMFPACK_Aat));
154: PetscFunctionReturn(PETSC_SUCCESS);
155: }
157: static PetscErrorCode MatSolveTranspose_UMFPACK(Mat A, Vec b, Vec x)
158: {
159: PetscFunctionBegin;
160: /* We gave UMFPACK the algebraic transpose (because it assumes column alignment) */
161: PetscCall(MatSolve_UMFPACK_Private(A, b, x, UMFPACK_A));
162: PetscFunctionReturn(PETSC_SUCCESS);
163: }
165: static PetscErrorCode MatLUFactorNumeric_UMFPACK(Mat F, Mat A, const MatFactorInfo *info)
166: {
167: Mat_UMFPACK *lu = (Mat_UMFPACK *)(F)->data;
168: Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data;
169: PetscInt *ai = a->i, *aj = a->j, status;
170: PetscScalar *av = a->a;
172: PetscFunctionBegin;
173: if (!A->rmap->n) PetscFunctionReturn(PETSC_SUCCESS);
174: /* numeric factorization of A' */
176: if (lu->flg == SAME_NONZERO_PATTERN && lu->Numeric) umfpack_UMF_free_numeric(&lu->Numeric);
177: #if defined(PETSC_USE_COMPLEX)
178: status = umfpack_UMF_numeric(ai, aj, (double *)av, NULL, lu->Symbolic, &lu->Numeric, lu->Control, lu->Info);
179: #else
180: status = umfpack_UMF_numeric(ai, aj, av, lu->Symbolic, &lu->Numeric, lu->Control, lu->Info);
181: #endif
182: if (status < 0) {
183: umfpack_UMF_report_status(lu->Control, status);
184: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "umfpack_UMF_numeric failed");
185: }
186: /* report numeric factorization of A' when Control[PRL] > 3 */
187: (void)umfpack_UMF_report_numeric(lu->Numeric, lu->Control);
189: PetscCall(PetscObjectReference((PetscObject)A));
190: PetscCall(MatDestroy(&lu->A));
192: lu->A = A;
193: lu->flg = SAME_NONZERO_PATTERN;
194: lu->CleanUpUMFPACK = PETSC_TRUE;
195: F->ops->solve = MatSolve_UMFPACK;
196: F->ops->solvetranspose = MatSolveTranspose_UMFPACK;
197: PetscFunctionReturn(PETSC_SUCCESS);
198: }
200: static PetscErrorCode MatLUFactorSymbolic_UMFPACK(Mat F, Mat A, IS r, IS c, const MatFactorInfo *info)
201: {
202: Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data;
203: Mat_UMFPACK *lu = (Mat_UMFPACK *)(F->data);
204: PetscInt i, *ai = a->i, *aj = a->j, m = A->rmap->n, n = A->cmap->n, status, idx;
205: #if !defined(PETSC_USE_COMPLEX)
206: PetscScalar *av = a->a;
207: #endif
208: const PetscInt *ra;
209: const char *strategy[] = {"AUTO", "UNSYMMETRIC", "SYMMETRIC"};
210: const char *scale[] = {"NONE", "SUM", "MAX"};
211: PetscBool flg;
213: PetscFunctionBegin;
214: (F)->ops->lufactornumeric = MatLUFactorNumeric_UMFPACK;
215: if (!n) PetscFunctionReturn(PETSC_SUCCESS);
217: /* Set options to F */
218: PetscOptionsBegin(PetscObjectComm((PetscObject)F), ((PetscObject)F)->prefix, "UMFPACK Options", "Mat");
219: /* Control parameters used by reporting routiones */
220: PetscCall(PetscOptionsReal("-mat_umfpack_prl", "Control[UMFPACK_PRL]", "None", lu->Control[UMFPACK_PRL], &lu->Control[UMFPACK_PRL], NULL));
222: /* Control parameters for symbolic factorization */
223: PetscCall(PetscOptionsEList("-mat_umfpack_strategy", "ordering and pivoting strategy", "None", strategy, 3, strategy[0], &idx, &flg));
224: if (flg) {
225: switch (idx) {
226: case 0:
227: lu->Control[UMFPACK_STRATEGY] = UMFPACK_STRATEGY_AUTO;
228: break;
229: case 1:
230: lu->Control[UMFPACK_STRATEGY] = UMFPACK_STRATEGY_UNSYMMETRIC;
231: break;
232: case 2:
233: lu->Control[UMFPACK_STRATEGY] = UMFPACK_STRATEGY_SYMMETRIC;
234: break;
235: }
236: }
237: PetscCall(PetscOptionsEList("-mat_umfpack_ordering", "Internal ordering method", "None", UmfpackOrderingTypes, PETSC_STATIC_ARRAY_LENGTH(UmfpackOrderingTypes), UmfpackOrderingTypes[(int)lu->Control[UMFPACK_ORDERING]], &idx, &flg));
238: if (flg) lu->Control[UMFPACK_ORDERING] = (int)idx;
239: PetscCall(PetscOptionsReal("-mat_umfpack_dense_col", "Control[UMFPACK_DENSE_COL]", "None", lu->Control[UMFPACK_DENSE_COL], &lu->Control[UMFPACK_DENSE_COL], NULL));
240: PetscCall(PetscOptionsReal("-mat_umfpack_dense_row", "Control[UMFPACK_DENSE_ROW]", "None", lu->Control[UMFPACK_DENSE_ROW], &lu->Control[UMFPACK_DENSE_ROW], NULL));
241: PetscCall(PetscOptionsReal("-mat_umfpack_amd_dense", "Control[UMFPACK_AMD_DENSE]", "None", lu->Control[UMFPACK_AMD_DENSE], &lu->Control[UMFPACK_AMD_DENSE], NULL));
242: PetscCall(PetscOptionsReal("-mat_umfpack_block_size", "Control[UMFPACK_BLOCK_SIZE]", "None", lu->Control[UMFPACK_BLOCK_SIZE], &lu->Control[UMFPACK_BLOCK_SIZE], NULL));
243: PetscCall(PetscOptionsReal("-mat_umfpack_fixq", "Control[UMFPACK_FIXQ]", "None", lu->Control[UMFPACK_FIXQ], &lu->Control[UMFPACK_FIXQ], NULL));
244: PetscCall(PetscOptionsReal("-mat_umfpack_aggressive", "Control[UMFPACK_AGGRESSIVE]", "None", lu->Control[UMFPACK_AGGRESSIVE], &lu->Control[UMFPACK_AGGRESSIVE], NULL));
246: /* Control parameters used by numeric factorization */
247: PetscCall(PetscOptionsReal("-mat_umfpack_pivot_tolerance", "Control[UMFPACK_PIVOT_TOLERANCE]", "None", lu->Control[UMFPACK_PIVOT_TOLERANCE], &lu->Control[UMFPACK_PIVOT_TOLERANCE], NULL));
248: PetscCall(PetscOptionsReal("-mat_umfpack_sym_pivot_tolerance", "Control[UMFPACK_SYM_PIVOT_TOLERANCE]", "None", lu->Control[UMFPACK_SYM_PIVOT_TOLERANCE], &lu->Control[UMFPACK_SYM_PIVOT_TOLERANCE], NULL));
249: PetscCall(PetscOptionsEList("-mat_umfpack_scale", "Control[UMFPACK_SCALE]", "None", scale, 3, scale[0], &idx, &flg));
250: if (flg) {
251: switch (idx) {
252: case 0:
253: lu->Control[UMFPACK_SCALE] = UMFPACK_SCALE_NONE;
254: break;
255: case 1:
256: lu->Control[UMFPACK_SCALE] = UMFPACK_SCALE_SUM;
257: break;
258: case 2:
259: lu->Control[UMFPACK_SCALE] = UMFPACK_SCALE_MAX;
260: break;
261: }
262: }
263: PetscCall(PetscOptionsReal("-mat_umfpack_alloc_init", "Control[UMFPACK_ALLOC_INIT]", "None", lu->Control[UMFPACK_ALLOC_INIT], &lu->Control[UMFPACK_ALLOC_INIT], NULL));
264: PetscCall(PetscOptionsReal("-mat_umfpack_front_alloc_init", "Control[UMFPACK_FRONT_ALLOC_INIT]", "None", lu->Control[UMFPACK_FRONT_ALLOC_INIT], &lu->Control[UMFPACK_ALLOC_INIT], NULL));
265: PetscCall(PetscOptionsReal("-mat_umfpack_droptol", "Control[UMFPACK_DROPTOL]", "None", lu->Control[UMFPACK_DROPTOL], &lu->Control[UMFPACK_DROPTOL], NULL));
267: /* Control parameters used by solve */
268: PetscCall(PetscOptionsReal("-mat_umfpack_irstep", "Control[UMFPACK_IRSTEP]", "None", lu->Control[UMFPACK_IRSTEP], &lu->Control[UMFPACK_IRSTEP], NULL));
269: PetscOptionsEnd();
271: if (r) {
272: PetscCall(ISGetIndices(r, &ra));
273: PetscCall(PetscMalloc1(m, &lu->perm_c));
274: /* we cannot simply memcpy on 64-bit archs */
275: for (i = 0; i < m; i++) lu->perm_c[i] = ra[i];
276: PetscCall(ISRestoreIndices(r, &ra));
277: }
279: /* print the control parameters */
280: if (lu->Control[UMFPACK_PRL] > 1) umfpack_UMF_report_control(lu->Control);
282: /* symbolic factorization of A' */
283: if (r) { /* use Petsc row ordering */
284: #if !defined(PETSC_USE_COMPLEX)
285: status = umfpack_UMF_qsymbolic(n, m, ai, aj, av, lu->perm_c, &lu->Symbolic, lu->Control, lu->Info);
286: #else
287: status = umfpack_UMF_qsymbolic(n, m, ai, aj, NULL, NULL, lu->perm_c, &lu->Symbolic, lu->Control, lu->Info);
288: #endif
289: } else { /* use Umfpack col ordering */
290: #if !defined(PETSC_USE_COMPLEX)
291: status = umfpack_UMF_symbolic(n, m, ai, aj, av, &lu->Symbolic, lu->Control, lu->Info);
292: #else
293: status = umfpack_UMF_symbolic(n, m, ai, aj, NULL, NULL, &lu->Symbolic, lu->Control, lu->Info);
294: #endif
295: }
296: if (status < 0) {
297: umfpack_UMF_report_info(lu->Control, lu->Info);
298: umfpack_UMF_report_status(lu->Control, status);
299: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "umfpack_UMF_symbolic failed");
300: }
301: /* report sumbolic factorization of A' when Control[PRL] > 3 */
302: (void)umfpack_UMF_report_symbolic(lu->Symbolic, lu->Control);
304: lu->flg = DIFFERENT_NONZERO_PATTERN;
305: lu->CleanUpUMFPACK = PETSC_TRUE;
306: PetscFunctionReturn(PETSC_SUCCESS);
307: }
309: static PetscErrorCode MatView_Info_UMFPACK(Mat A, PetscViewer viewer)
310: {
311: Mat_UMFPACK *lu = (Mat_UMFPACK *)A->data;
313: PetscFunctionBegin;
314: /* check if matrix is UMFPACK type */
315: if (A->ops->solve != MatSolve_UMFPACK) PetscFunctionReturn(PETSC_SUCCESS);
317: PetscCall(PetscViewerASCIIPrintf(viewer, "UMFPACK run parameters:\n"));
318: /* Control parameters used by reporting routiones */
319: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_PRL]: %g\n", lu->Control[UMFPACK_PRL]));
321: /* Control parameters used by symbolic factorization */
322: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_STRATEGY]: %g\n", lu->Control[UMFPACK_STRATEGY]));
323: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_DENSE_COL]: %g\n", lu->Control[UMFPACK_DENSE_COL]));
324: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_DENSE_ROW]: %g\n", lu->Control[UMFPACK_DENSE_ROW]));
325: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_AMD_DENSE]: %g\n", lu->Control[UMFPACK_AMD_DENSE]));
326: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_BLOCK_SIZE]: %g\n", lu->Control[UMFPACK_BLOCK_SIZE]));
327: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_FIXQ]: %g\n", lu->Control[UMFPACK_FIXQ]));
328: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_AGGRESSIVE]: %g\n", lu->Control[UMFPACK_AGGRESSIVE]));
330: /* Control parameters used by numeric factorization */
331: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_PIVOT_TOLERANCE]: %g\n", lu->Control[UMFPACK_PIVOT_TOLERANCE]));
332: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_SYM_PIVOT_TOLERANCE]: %g\n", lu->Control[UMFPACK_SYM_PIVOT_TOLERANCE]));
333: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_SCALE]: %g\n", lu->Control[UMFPACK_SCALE]));
334: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_ALLOC_INIT]: %g\n", lu->Control[UMFPACK_ALLOC_INIT]));
335: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_DROPTOL]: %g\n", lu->Control[UMFPACK_DROPTOL]));
337: /* Control parameters used by solve */
338: PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_IRSTEP]: %g\n", lu->Control[UMFPACK_IRSTEP]));
340: /* mat ordering */
341: if (!lu->perm_c) PetscCall(PetscViewerASCIIPrintf(viewer, " Control[UMFPACK_ORDERING]: %s (not using the PETSc ordering)\n", UmfpackOrderingTypes[(int)lu->Control[UMFPACK_ORDERING]]));
342: PetscFunctionReturn(PETSC_SUCCESS);
343: }
345: static PetscErrorCode MatView_UMFPACK(Mat A, PetscViewer viewer)
346: {
347: PetscBool iascii;
348: PetscViewerFormat format;
350: PetscFunctionBegin;
351: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
352: if (iascii) {
353: PetscCall(PetscViewerGetFormat(viewer, &format));
354: if (format == PETSC_VIEWER_ASCII_INFO) PetscCall(MatView_Info_UMFPACK(A, viewer));
355: }
356: PetscFunctionReturn(PETSC_SUCCESS);
357: }
359: PetscErrorCode MatFactorGetSolverType_seqaij_umfpack(Mat A, MatSolverType *type)
360: {
361: PetscFunctionBegin;
362: *type = MATSOLVERUMFPACK;
363: PetscFunctionReturn(PETSC_SUCCESS);
364: }
366: /*MC
367: MATSOLVERUMFPACK = "umfpack" - A matrix type providing direct solvers, LU, for sequential matrices
368: via the external package UMFPACK.
370: Use `./configure --download-suitesparse` to install PETSc to use UMFPACK
372: Use `-pc_type lu` `-pc_factor_mat_solver_type umfpack` to use this direct solver
374: Consult UMFPACK documentation for more information about the Control parameters
375: which correspond to the options database keys below.
377: Options Database Keys:
378: + -mat_umfpack_ordering - `CHOLMOD`, `AMD`, `GIVEN`, `METIS`, `BEST`, `NONE`
379: . -mat_umfpack_prl - UMFPACK print level: Control[UMFPACK_PRL]
380: . -mat_umfpack_strategy <AUTO> - (choose one of) `AUTO`, `UNSYMMETRIC`, `SYMMETRIC 2BY2`
381: . -mat_umfpack_dense_col <alpha_c> - UMFPACK dense column threshold: Control[UMFPACK_DENSE_COL]
382: . -mat_umfpack_dense_row <0.2> - Control[UMFPACK_DENSE_ROW]
383: . -mat_umfpack_amd_dense <10> - Control[UMFPACK_AMD_DENSE]
384: . -mat_umfpack_block_size <bs> - UMFPACK block size for BLAS-Level 3 calls: Control[UMFPACK_BLOCK_SIZE]
385: . -mat_umfpack_2by2_tolerance <0.01> - Control[UMFPACK_2BY2_TOLERANCE]
386: . -mat_umfpack_fixq <0> - Control[UMFPACK_FIXQ]
387: . -mat_umfpack_aggressive <1> - Control[UMFPACK_AGGRESSIVE]
388: . -mat_umfpack_pivot_tolerance <delta> - UMFPACK partial pivot tolerance: Control[UMFPACK_PIVOT_TOLERANCE]
389: . -mat_umfpack_sym_pivot_tolerance <0.001> - Control[UMFPACK_SYM_PIVOT_TOLERANCE]
390: . -mat_umfpack_scale <NONE> - (choose one of) NONE SUM MAX
391: . -mat_umfpack_alloc_init <delta> - UMFPACK factorized matrix allocation modifier: Control[UMFPACK_ALLOC_INIT]
392: . -mat_umfpack_droptol <0> - Control[UMFPACK_DROPTOL]
393: - -mat_umfpack_irstep <maxit> - UMFPACK maximum number of iterative refinement steps: Control[UMFPACK_IRSTEP]
395: Level: beginner
397: Note:
398: UMFPACK is part of SuiteSparse http://faculty.cse.tamu.edu/davis/suitesparse.html
400: .seealso: [](ch_matrices), `Mat`, `PCLU`, `MATSOLVERSUPERLU`, `MATSOLVERMUMPS`, `PCFactorSetMatSolverType()`, `MatSolverType`
401: M*/
403: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_umfpack(Mat A, MatFactorType ftype, Mat *F)
404: {
405: Mat B;
406: Mat_UMFPACK *lu;
407: PetscInt m = A->rmap->n, n = A->cmap->n;
409: PetscFunctionBegin;
410: /* Create the factorization matrix F */
411: PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
412: PetscCall(MatSetSizes(B, PETSC_DECIDE, PETSC_DECIDE, m, n));
413: PetscCall(PetscStrallocpy("umfpack", &((PetscObject)B)->type_name));
414: PetscCall(MatSetUp(B));
416: PetscCall(PetscNew(&lu));
418: B->data = lu;
419: B->ops->getinfo = MatGetInfo_External;
420: B->ops->lufactorsymbolic = MatLUFactorSymbolic_UMFPACK;
421: B->ops->destroy = MatDestroy_UMFPACK;
422: B->ops->view = MatView_UMFPACK;
423: B->ops->matsolve = NULL;
425: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_seqaij_umfpack));
427: B->factortype = MAT_FACTOR_LU;
428: B->assembled = PETSC_TRUE; /* required by -ksp_view */
429: B->preallocated = PETSC_TRUE;
431: PetscCall(PetscFree(B->solvertype));
432: PetscCall(PetscStrallocpy(MATSOLVERUMFPACK, &B->solvertype));
433: B->canuseordering = PETSC_TRUE;
434: PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL, (char **)&B->preferredordering[MAT_FACTOR_LU]));
436: /* initializations */
437: /* get the default control parameters */
438: umfpack_UMF_defaults(lu->Control);
439: lu->perm_c = NULL; /* use default UMFPACK col permutation */
440: lu->Control[UMFPACK_IRSTEP] = 0; /* max num of iterative refinement steps to attempt */
442: *F = B;
443: PetscFunctionReturn(PETSC_SUCCESS);
444: }
446: PETSC_INTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat, MatFactorType, Mat *);
447: PETSC_INTERN PetscErrorCode MatGetFactor_seqsbaij_cholmod(Mat, MatFactorType, Mat *);
448: PETSC_INTERN PetscErrorCode MatGetFactor_seqaij_klu(Mat, MatFactorType, Mat *);
449: PETSC_INTERN PetscErrorCode MatGetFactor_seqaij_spqr(Mat, MatFactorType, Mat *);
451: PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_SuiteSparse(void)
452: {
453: PetscFunctionBegin;
454: PetscCall(MatSolverTypeRegister(MATSOLVERUMFPACK, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_umfpack));
455: PetscCall(MatSolverTypeRegister(MATSOLVERCHOLMOD, MATSEQAIJ, MAT_FACTOR_CHOLESKY, MatGetFactor_seqaij_cholmod));
456: PetscCall(MatSolverTypeRegister(MATSOLVERCHOLMOD, MATSEQSBAIJ, MAT_FACTOR_CHOLESKY, MatGetFactor_seqsbaij_cholmod));
457: PetscCall(MatSolverTypeRegister(MATSOLVERKLU, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_klu));
458: PetscCall(MatSolverTypeRegister(MATSOLVERSPQR, MATSEQAIJ, MAT_FACTOR_QR, MatGetFactor_seqaij_spqr));
459: if (!PetscDefined(USE_COMPLEX)) PetscCall(MatSolverTypeRegister(MATSOLVERSPQR, MATNORMAL, MAT_FACTOR_QR, MatGetFactor_seqaij_spqr));
460: PetscCall(MatSolverTypeRegister(MATSOLVERSPQR, MATNORMALHERMITIAN, MAT_FACTOR_QR, MatGetFactor_seqaij_spqr));
461: PetscFunctionReturn(PETSC_SUCCESS);
462: }