Actual source code: baijsolvtrannat4.c
1: #include <../src/mat/impls/baij/seq/baij.h>
3: PetscErrorCode MatSolveTranspose_SeqBAIJ_4_NaturalOrdering_inplace(Mat A, Vec bb, Vec xx)
4: {
5: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
6: const PetscInt *diag = a->diag, n = a->mbs, *vi, *ai = a->i, *aj = a->j;
7: PetscInt i, nz, idx, idt, oidx;
8: const MatScalar *aa = a->a, *v;
9: PetscScalar s1, s2, s3, s4, x1, x2, x3, x4, *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 + 16 * 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: x4 = x[3 + idx];
23: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3 + v[3] * x4;
24: s2 = v[4] * x1 + v[5] * x2 + v[6] * x3 + v[7] * x4;
25: s3 = v[8] * x1 + v[9] * x2 + v[10] * x3 + v[11] * x4;
26: s4 = v[12] * x1 + v[13] * x2 + v[14] * x3 + v[15] * x4;
27: v += 16;
29: vi = aj + diag[i] + 1;
30: nz = ai[i + 1] - diag[i] - 1;
31: while (nz--) {
32: oidx = 4 * (*vi++);
33: x[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4;
34: x[oidx + 1] -= v[4] * s1 + v[5] * s2 + v[6] * s3 + v[7] * s4;
35: x[oidx + 2] -= v[8] * s1 + v[9] * s2 + v[10] * s3 + v[11] * s4;
36: x[oidx + 3] -= v[12] * s1 + v[13] * s2 + v[14] * s3 + v[15] * s4;
37: v += 16;
38: }
39: x[idx] = s1;
40: x[1 + idx] = s2;
41: x[2 + idx] = s3;
42: x[3 + idx] = s4;
43: idx += 4;
44: }
45: /* backward solve the L^T */
46: for (i = n - 1; i >= 0; i--) {
47: v = aa + 16 * diag[i] - 16;
48: vi = aj + diag[i] - 1;
49: nz = diag[i] - ai[i];
50: idt = 4 * i;
51: s1 = x[idt];
52: s2 = x[1 + idt];
53: s3 = x[2 + idt];
54: s4 = x[3 + idt];
55: while (nz--) {
56: idx = 4 * (*vi--);
57: x[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4;
58: x[idx + 1] -= v[4] * s1 + v[5] * s2 + v[6] * s3 + v[7] * s4;
59: x[idx + 2] -= v[8] * s1 + v[9] * s2 + v[10] * s3 + v[11] * s4;
60: x[idx + 3] -= v[12] * s1 + v[13] * s2 + v[14] * s3 + v[15] * s4;
61: v -= 16;
62: }
63: }
64: VecRestoreArray(xx, &x);
65: PetscLogFlops(2.0 * 16 * (a->nz) - 4.0 * A->cmap->n);
66: return 0;
67: }
69: PetscErrorCode MatSolveTranspose_SeqBAIJ_4_NaturalOrdering(Mat A, Vec bb, Vec xx)
70: {
71: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
72: const PetscInt n = a->mbs, *vi, *ai = a->i, *aj = a->j, *diag = a->diag;
73: PetscInt nz, idx, idt, j, i, oidx;
74: const PetscInt bs = A->rmap->bs, bs2 = a->bs2;
75: const MatScalar *aa = a->a, *v;
76: PetscScalar s1, s2, s3, s4, x1, x2, x3, x4, *x;
78: VecCopy(bb, xx);
79: VecGetArray(xx, &x);
81: /* forward solve the U^T */
82: idx = 0;
83: for (i = 0; i < n; i++) {
84: v = aa + bs2 * diag[i];
85: /* multiply by the inverse of the block diagonal */
86: x1 = x[idx];
87: x2 = x[1 + idx];
88: x3 = x[2 + idx];
89: x4 = x[3 + idx];
90: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3 + v[3] * x4;
91: s2 = v[4] * x1 + v[5] * x2 + v[6] * x3 + v[7] * x4;
92: s3 = v[8] * x1 + v[9] * x2 + v[10] * x3 + v[11] * x4;
93: s4 = v[12] * x1 + v[13] * x2 + v[14] * x3 + v[15] * x4;
94: v -= bs2;
96: vi = aj + diag[i] - 1;
97: nz = diag[i] - diag[i + 1] - 1;
98: for (j = 0; j > -nz; j--) {
99: oidx = bs * vi[j];
100: x[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4;
101: x[oidx + 1] -= v[4] * s1 + v[5] * s2 + v[6] * s3 + v[7] * s4;
102: x[oidx + 2] -= v[8] * s1 + v[9] * s2 + v[10] * s3 + v[11] * s4;
103: x[oidx + 3] -= v[12] * s1 + v[13] * s2 + v[14] * s3 + v[15] * s4;
104: v -= bs2;
105: }
106: x[idx] = s1;
107: x[1 + idx] = s2;
108: x[2 + idx] = s3;
109: x[3 + idx] = s4;
110: idx += bs;
111: }
112: /* backward solve the L^T */
113: for (i = n - 1; i >= 0; i--) {
114: v = aa + bs2 * ai[i];
115: vi = aj + ai[i];
116: nz = ai[i + 1] - ai[i];
117: idt = bs * i;
118: s1 = x[idt];
119: s2 = x[1 + idt];
120: s3 = x[2 + idt];
121: s4 = x[3 + idt];
122: for (j = 0; j < nz; j++) {
123: idx = bs * vi[j];
124: x[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4;
125: x[idx + 1] -= v[4] * s1 + v[5] * s2 + v[6] * s3 + v[7] * s4;
126: x[idx + 2] -= v[8] * s1 + v[9] * s2 + v[10] * s3 + v[11] * s4;
127: x[idx + 3] -= v[12] * s1 + v[13] * s2 + v[14] * s3 + v[15] * s4;
128: v += bs2;
129: }
130: }
131: VecRestoreArray(xx, &x);
132: PetscLogFlops(2.0 * bs2 * (a->nz) - bs * A->cmap->n);
133: return 0;
134: }