Actual source code: baijfact4.c
1: /*
2: Factorization code for BAIJ format.
3: */
4: #include <../src/mat/impls/baij/seq/baij.h>
5: #include <petsc/private/kernels/blockinvert.h>
7: PetscErrorCode MatLUFactorNumeric_SeqBAIJ_N_inplace(Mat C, Mat A, const MatFactorInfo *info)
8: {
9: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data, *b = (Mat_SeqBAIJ *)C->data;
10: IS isrow = b->row, isicol = b->icol;
11: const PetscInt *r, *ic;
12: PetscInt i, j, n = a->mbs, *bi = b->i, *bj = b->j;
13: PetscInt *ajtmpold, *ajtmp, nz, row, *ai = a->i, *aj = a->j, k, flg;
14: PetscInt *diag_offset = b->diag, diag, bs = A->rmap->bs, bs2 = a->bs2, *pj, *v_pivots;
15: MatScalar *ba = b->a, *aa = a->a, *pv, *v, *rtmp, *multiplier, *v_work, *pc, *w;
16: PetscBool allowzeropivot, zeropivotdetected;
18: PetscFunctionBegin;
19: PetscCall(ISGetIndices(isrow, &r));
20: PetscCall(ISGetIndices(isicol, &ic));
21: allowzeropivot = PetscNot(A->erroriffailure);
23: PetscCall(PetscCalloc1(bs2 * (n + 1), &rtmp));
24: /* generate work space needed by dense LU factorization */
25: PetscCall(PetscMalloc3(bs, &v_work, bs2, &multiplier, bs, &v_pivots));
27: for (i = 0; i < n; i++) {
28: nz = bi[i + 1] - bi[i];
29: ajtmp = bj + bi[i];
30: for (j = 0; j < nz; j++) PetscCall(PetscArrayzero(rtmp + bs2 * ajtmp[j], bs2));
31: /* load in initial (unfactored row) */
32: nz = ai[r[i] + 1] - ai[r[i]];
33: ajtmpold = aj + ai[r[i]];
34: v = aa + bs2 * ai[r[i]];
35: for (j = 0; j < nz; j++) PetscCall(PetscArraycpy(rtmp + bs2 * ic[ajtmpold[j]], v + bs2 * j, bs2));
36: row = *ajtmp++;
37: while (row < i) {
38: pc = rtmp + bs2 * row;
39: /* if (*pc) { */
40: for (flg = 0, k = 0; k < bs2; k++) {
41: if (pc[k] != 0.0) {
42: flg = 1;
43: break;
44: }
45: }
46: if (flg) {
47: pv = ba + bs2 * diag_offset[row];
48: pj = bj + diag_offset[row] + 1;
49: PetscKernel_A_gets_A_times_B(bs, pc, pv, multiplier);
50: nz = bi[row + 1] - diag_offset[row] - 1;
51: pv += bs2;
52: for (j = 0; j < nz; j++) PetscKernel_A_gets_A_minus_B_times_C(bs, rtmp + bs2 * pj[j], pc, pv + bs2 * j);
53: PetscCall(PetscLogFlops(2.0 * bs * bs2 * (nz + 1.0) - bs));
54: }
55: row = *ajtmp++;
56: }
57: /* finished row so stick it into b->a */
58: pv = ba + bs2 * bi[i];
59: pj = bj + bi[i];
60: nz = bi[i + 1] - bi[i];
61: for (j = 0; j < nz; j++) PetscCall(PetscArraycpy(pv + bs2 * j, rtmp + bs2 * pj[j], bs2));
62: diag = diag_offset[i] - bi[i];
63: /* invert diagonal block */
64: w = pv + bs2 * diag;
66: PetscCall(PetscKernel_A_gets_inverse_A(bs, w, v_pivots, v_work, allowzeropivot, &zeropivotdetected));
67: if (zeropivotdetected) C->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
68: }
70: PetscCall(PetscFree(rtmp));
71: PetscCall(PetscFree3(v_work, multiplier, v_pivots));
72: PetscCall(ISRestoreIndices(isicol, &ic));
73: PetscCall(ISRestoreIndices(isrow, &r));
75: C->ops->solve = MatSolve_SeqBAIJ_N_inplace;
76: C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_N_inplace;
77: C->assembled = PETSC_TRUE;
79: PetscCall(PetscLogFlops(1.333333333333 * bs * bs2 * b->mbs)); /* from inverting diagonal blocks */
80: PetscFunctionReturn(PETSC_SUCCESS);
81: }