Actual source code: baijsolvtrannat3.c
1: #include <../src/mat/impls/baij/seq/baij.h>
3: PetscErrorCode MatSolveTranspose_SeqBAIJ_3_NaturalOrdering_inplace(Mat A, Vec bb, Vec xx)
4: {
5: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
6: const PetscInt n = a->mbs, *vi, *ai = a->i, *aj = a->j, *diag = a->diag;
7: PetscInt i, nz, idx, idt, oidx;
8: const MatScalar *aa = a->a, *v;
9: PetscScalar s1, s2, s3, x1, x2, x3, *x;
11: VecCopy(bb, xx);
12: VecGetArray(xx, &x);
14: /* forward solve the U^T */
15: idx = 0;
16: for (i = 0; i < n; i++) {
17: v = aa + 9 * diag[i];
18: /* multiply by the inverse of the block diagonal */
19: x1 = x[idx];
20: x2 = x[1 + idx];
21: x3 = x[2 + idx];
22: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3;
23: s2 = v[3] * x1 + v[4] * x2 + v[5] * x3;
24: s3 = v[6] * x1 + v[7] * x2 + v[8] * x3;
25: v += 9;
27: vi = aj + diag[i] + 1;
28: nz = ai[i + 1] - diag[i] - 1;
29: while (nz--) {
30: oidx = 3 * (*vi++);
31: x[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3;
32: x[oidx + 1] -= v[3] * s1 + v[4] * s2 + v[5] * s3;
33: x[oidx + 2] -= v[6] * s1 + v[7] * s2 + v[8] * s3;
34: v += 9;
35: }
36: x[idx] = s1;
37: x[1 + idx] = s2;
38: x[2 + idx] = s3;
39: idx += 3;
40: }
41: /* backward solve the L^T */
42: for (i = n - 1; i >= 0; i--) {
43: v = aa + 9 * diag[i] - 9;
44: vi = aj + diag[i] - 1;
45: nz = diag[i] - ai[i];
46: idt = 3 * i;
47: s1 = x[idt];
48: s2 = x[1 + idt];
49: s3 = x[2 + idt];
50: while (nz--) {
51: idx = 3 * (*vi--);
52: x[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3;
53: x[idx + 1] -= v[3] * s1 + v[4] * s2 + v[5] * s3;
54: x[idx + 2] -= v[6] * s1 + v[7] * s2 + v[8] * s3;
55: v -= 9;
56: }
57: }
58: VecRestoreArray(xx, &x);
59: PetscLogFlops(2.0 * 9.0 * (a->nz) - 3.0 * A->cmap->n);
60: return 0;
61: }
63: PetscErrorCode MatSolveTranspose_SeqBAIJ_3_NaturalOrdering(Mat A, Vec bb, Vec xx)
64: {
65: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
66: const PetscInt n = a->mbs, *vi, *ai = a->i, *aj = a->j, *diag = a->diag;
67: PetscInt nz, idx, idt, j, i, oidx;
68: const PetscInt bs = A->rmap->bs, bs2 = a->bs2;
69: const MatScalar *aa = a->a, *v;
70: PetscScalar s1, s2, s3, x1, x2, x3, *x;
72: VecCopy(bb, xx);
73: VecGetArray(xx, &x);
75: /* forward solve the U^T */
76: idx = 0;
77: for (i = 0; i < n; i++) {
78: v = aa + bs2 * diag[i];
79: /* multiply by the inverse of the block diagonal */
80: x1 = x[idx];
81: x2 = x[1 + idx];
82: x3 = x[2 + idx];
83: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3;
84: s2 = v[3] * x1 + v[4] * x2 + v[5] * x3;
85: s3 = v[6] * x1 + v[7] * x2 + v[8] * x3;
86: v -= bs2;
88: vi = aj + diag[i] - 1;
89: nz = diag[i] - diag[i + 1] - 1;
90: for (j = 0; j > -nz; j--) {
91: oidx = bs * vi[j];
92: x[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3;
93: x[oidx + 1] -= v[3] * s1 + v[4] * s2 + v[5] * s3;
94: x[oidx + 2] -= v[6] * s1 + v[7] * s2 + v[8] * s3;
95: v -= bs2;
96: }
97: x[idx] = s1;
98: x[1 + idx] = s2;
99: x[2 + idx] = s3;
100: idx += bs;
101: }
102: /* backward solve the L^T */
103: for (i = n - 1; i >= 0; i--) {
104: v = aa + bs2 * ai[i];
105: vi = aj + ai[i];
106: nz = ai[i + 1] - ai[i];
107: idt = bs * i;
108: s1 = x[idt];
109: s2 = x[1 + idt];
110: s3 = x[2 + idt];
111: for (j = 0; j < nz; j++) {
112: idx = bs * vi[j];
113: x[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3;
114: x[idx + 1] -= v[3] * s1 + v[4] * s2 + v[5] * s3;
115: x[idx + 2] -= v[6] * s1 + v[7] * s2 + v[8] * s3;
116: v += bs2;
117: }
118: }
119: VecRestoreArray(xx, &x);
120: PetscLogFlops(2.0 * bs2 * (a->nz) - bs * A->cmap->n);
121: return 0;
122: }