Actual source code: aijcholmod.c
2: #include <../src/mat/impls/aij/seq/aij.h>
3: #include <../src/mat/impls/sbaij/seq/cholmod/cholmodimpl.h>
5: static PetscErrorCode MatWrapCholmod_seqaij(Mat A, PetscBool values, cholmod_sparse *C, PetscBool *aijalloc, PetscBool *valloc)
6: {
7: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)A->data;
8: const PetscScalar *aa;
9: PetscScalar *ca;
10: const PetscInt *ai = aij->i, *aj = aij->j, *adiag;
11: PetscInt m = A->rmap->n, i, j, k, nz, *ci, *cj;
12: PetscBool vain = PETSC_FALSE;
14: PetscFunctionBegin;
15: PetscCall(MatMarkDiagonal_SeqAIJ(A));
16: adiag = aij->diag;
17: for (i = 0, nz = 0; i < m; i++) nz += ai[i + 1] - adiag[i];
18: PetscCall(PetscMalloc2(m + 1, &ci, nz, &cj));
19: if (values) {
20: vain = PETSC_TRUE;
21: PetscCall(PetscMalloc1(nz, &ca));
22: PetscCall(MatSeqAIJGetArrayRead(A, &aa));
23: }
24: for (i = 0, k = 0; i < m; i++) {
25: ci[i] = k;
26: for (j = adiag[i]; j < ai[i + 1]; j++, k++) {
27: cj[k] = aj[j];
28: if (values) ca[k] = PetscConj(aa[j]);
29: }
30: }
31: ci[i] = k;
32: *aijalloc = PETSC_TRUE;
33: *valloc = vain;
34: if (values) PetscCall(MatSeqAIJRestoreArrayRead(A, &aa));
36: PetscCall(PetscMemzero(C, sizeof(*C)));
38: C->nrow = (size_t)A->cmap->n;
39: C->ncol = (size_t)A->rmap->n;
40: C->nzmax = (size_t)nz;
41: C->p = ci;
42: C->i = cj;
43: C->x = values ? ca : 0;
44: C->stype = -1;
45: C->itype = CHOLMOD_INT_TYPE;
46: C->xtype = values ? CHOLMOD_SCALAR_TYPE : CHOLMOD_PATTERN;
47: C->dtype = CHOLMOD_DOUBLE;
48: C->sorted = 1;
49: C->packed = 1;
50: PetscFunctionReturn(PETSC_SUCCESS);
51: }
53: static PetscErrorCode MatFactorGetSolverType_seqaij_cholmod(Mat A, MatSolverType *type)
54: {
55: PetscFunctionBegin;
56: *type = MATSOLVERCHOLMOD;
57: PetscFunctionReturn(PETSC_SUCCESS);
58: }
60: /* Almost a copy of MatGetFactor_seqsbaij_cholmod, yuck */
61: PETSC_INTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat A, MatFactorType ftype, Mat *F)
62: {
63: Mat B;
64: Mat_CHOLMOD *chol;
65: PetscInt m = A->rmap->n, n = A->cmap->n;
67: PetscFunctionBegin;
68: #if defined(PETSC_USE_COMPLEX)
69: PetscCheck(A->hermitian == PETSC_BOOL3_TRUE, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Only for Hermitian matrices");
70: #endif
71: /* Create the factorization matrix F */
72: PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
73: PetscCall(MatSetSizes(B, PETSC_DECIDE, PETSC_DECIDE, m, n));
74: PetscCall(PetscStrallocpy("cholmod", &((PetscObject)B)->type_name));
75: PetscCall(MatSetUp(B));
76: PetscCall(PetscNew(&chol));
78: chol->Wrap = MatWrapCholmod_seqaij;
79: B->data = chol;
81: B->ops->getinfo = MatGetInfo_CHOLMOD;
82: B->ops->view = MatView_CHOLMOD;
83: B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_CHOLMOD;
84: B->ops->destroy = MatDestroy_CHOLMOD;
86: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_seqaij_cholmod));
88: B->factortype = MAT_FACTOR_CHOLESKY;
89: B->assembled = PETSC_TRUE;
90: B->preallocated = PETSC_TRUE;
92: PetscCall(PetscFree(B->solvertype));
93: PetscCall(PetscStrallocpy(MATSOLVERCHOLMOD, &B->solvertype));
94: B->canuseordering = PETSC_TRUE;
95: PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL, (char **)&B->preferredordering[MAT_FACTOR_CHOLESKY]));
96: PetscCall(CholmodStart(B));
97: *F = B;
98: PetscFunctionReturn(PETSC_SUCCESS);
99: }