Actual source code: baijsolvtrannat7.c
1: #include <../src/mat/impls/baij/seq/baij.h>
3: PetscErrorCode MatSolveTranspose_SeqBAIJ_7_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, s5, s6, s7, x1, x2, x3, x4, x5, x6, x7, *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 + 49 * 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: x5 = x[4 + idx];
24: x6 = x[5 + idx];
25: x7 = x[6 + idx];
26: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3 + v[3] * x4 + v[4] * x5 + v[5] * x6 + v[6] * x7;
27: s2 = v[7] * x1 + v[8] * x2 + v[9] * x3 + v[10] * x4 + v[11] * x5 + v[12] * x6 + v[13] * x7;
28: s3 = v[14] * x1 + v[15] * x2 + v[16] * x3 + v[17] * x4 + v[18] * x5 + v[19] * x6 + v[20] * x7;
29: s4 = v[21] * x1 + v[22] * x2 + v[23] * x3 + v[24] * x4 + v[25] * x5 + v[26] * x6 + v[27] * x7;
30: s5 = v[28] * x1 + v[29] * x2 + v[30] * x3 + v[31] * x4 + v[32] * x5 + v[33] * x6 + v[34] * x7;
31: s6 = v[35] * x1 + v[36] * x2 + v[37] * x3 + v[38] * x4 + v[39] * x5 + v[40] * x6 + v[41] * x7;
32: s7 = v[42] * x1 + v[43] * x2 + v[44] * x3 + v[45] * x4 + v[46] * x5 + v[47] * x6 + v[48] * x7;
33: v += 49;
35: vi = aj + diag[i] + 1;
36: nz = ai[i + 1] - diag[i] - 1;
37: while (nz--) {
38: oidx = 7 * (*vi++);
39: x[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5 + v[5] * s6 + v[6] * s7;
40: x[oidx + 1] -= v[7] * s1 + v[8] * s2 + v[9] * s3 + v[10] * s4 + v[11] * s5 + v[12] * s6 + v[13] * s7;
41: x[oidx + 2] -= v[14] * s1 + v[15] * s2 + v[16] * s3 + v[17] * s4 + v[18] * s5 + v[19] * s6 + v[20] * s7;
42: x[oidx + 3] -= v[21] * s1 + v[22] * s2 + v[23] * s3 + v[24] * s4 + v[25] * s5 + v[26] * s6 + v[27] * s7;
43: x[oidx + 4] -= v[28] * s1 + v[29] * s2 + v[30] * s3 + v[31] * s4 + v[32] * s5 + v[33] * s6 + v[34] * s7;
44: x[oidx + 5] -= v[35] * s1 + v[36] * s2 + v[37] * s3 + v[38] * s4 + v[39] * s5 + v[40] * s6 + v[41] * s7;
45: x[oidx + 6] -= v[42] * s1 + v[43] * s2 + v[44] * s3 + v[45] * s4 + v[46] * s5 + v[47] * s6 + v[48] * s7;
46: v += 49;
47: }
48: x[idx] = s1;
49: x[1 + idx] = s2;
50: x[2 + idx] = s3;
51: x[3 + idx] = s4;
52: x[4 + idx] = s5;
53: x[5 + idx] = s6;
54: x[6 + idx] = s7;
55: idx += 7;
56: }
57: /* backward solve the L^T */
58: for (i = n - 1; i >= 0; i--) {
59: v = aa + 49 * diag[i] - 49;
60: vi = aj + diag[i] - 1;
61: nz = diag[i] - ai[i];
62: idt = 7 * i;
63: s1 = x[idt];
64: s2 = x[1 + idt];
65: s3 = x[2 + idt];
66: s4 = x[3 + idt];
67: s5 = x[4 + idt];
68: s6 = x[5 + idt];
69: s7 = x[6 + idt];
70: while (nz--) {
71: idx = 7 * (*vi--);
72: x[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5 + v[5] * s6 + v[6] * s7;
73: x[idx + 1] -= v[7] * s1 + v[8] * s2 + v[9] * s3 + v[10] * s4 + v[11] * s5 + v[12] * s6 + v[13] * s7;
74: x[idx + 2] -= v[14] * s1 + v[15] * s2 + v[16] * s3 + v[17] * s4 + v[18] * s5 + v[19] * s6 + v[20] * s7;
75: x[idx + 3] -= v[21] * s1 + v[22] * s2 + v[23] * s3 + v[24] * s4 + v[25] * s5 + v[26] * s6 + v[27] * s7;
76: x[idx + 4] -= v[28] * s1 + v[29] * s2 + v[30] * s3 + v[31] * s4 + v[32] * s5 + v[33] * s6 + v[34] * s7;
77: x[idx + 5] -= v[35] * s1 + v[36] * s2 + v[37] * s3 + v[38] * s4 + v[39] * s5 + v[40] * s6 + v[41] * s7;
78: x[idx + 6] -= v[42] * s1 + v[43] * s2 + v[44] * s3 + v[45] * s4 + v[46] * s5 + v[47] * s6 + v[48] * s7;
79: v -= 49;
80: }
81: }
82: VecRestoreArray(xx, &x);
83: PetscLogFlops(2.0 * 49 * (a->nz) - 7.0 * A->cmap->n);
84: return 0;
85: }
86: PetscErrorCode MatSolveTranspose_SeqBAIJ_7_NaturalOrdering(Mat A, Vec bb, Vec xx)
87: {
88: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
89: const PetscInt n = a->mbs, *vi, *ai = a->i, *aj = a->j, *diag = a->diag;
90: PetscInt nz, idx, idt, j, i, oidx;
91: const PetscInt bs = A->rmap->bs, bs2 = a->bs2;
92: const MatScalar *aa = a->a, *v;
93: PetscScalar s1, s2, s3, s4, s5, s6, s7, x1, x2, x3, x4, x5, x6, x7, *x;
95: VecCopy(bb, xx);
96: VecGetArray(xx, &x);
98: /* forward solve the U^T */
99: idx = 0;
100: for (i = 0; i < n; i++) {
101: v = aa + bs2 * diag[i];
102: /* multiply by the inverse of the block diagonal */
103: x1 = x[idx];
104: x2 = x[1 + idx];
105: x3 = x[2 + idx];
106: x4 = x[3 + idx];
107: x5 = x[4 + idx];
108: x6 = x[5 + idx];
109: x7 = x[6 + idx];
110: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3 + v[3] * x4 + v[4] * x5 + v[5] * x6 + v[6] * x7;
111: s2 = v[7] * x1 + v[8] * x2 + v[9] * x3 + v[10] * x4 + v[11] * x5 + v[12] * x6 + v[13] * x7;
112: s3 = v[14] * x1 + v[15] * x2 + v[16] * x3 + v[17] * x4 + v[18] * x5 + v[19] * x6 + v[20] * x7;
113: s4 = v[21] * x1 + v[22] * x2 + v[23] * x3 + v[24] * x4 + v[25] * x5 + v[26] * x6 + v[27] * x7;
114: s5 = v[28] * x1 + v[29] * x2 + v[30] * x3 + v[31] * x4 + v[32] * x5 + v[33] * x6 + v[34] * x7;
115: s6 = v[35] * x1 + v[36] * x2 + v[37] * x3 + v[38] * x4 + v[39] * x5 + v[40] * x6 + v[41] * x7;
116: s7 = v[42] * x1 + v[43] * x2 + v[44] * x3 + v[45] * x4 + v[46] * x5 + v[47] * x6 + v[48] * x7;
117: v -= bs2;
118: vi = aj + diag[i] - 1;
119: nz = diag[i] - diag[i + 1] - 1;
120: for (j = 0; j > -nz; j--) {
121: oidx = bs * vi[j];
122: x[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5 + v[5] * s6 + v[6] * s7;
123: x[oidx + 1] -= v[7] * s1 + v[8] * s2 + v[9] * s3 + v[10] * s4 + v[11] * s5 + v[12] * s6 + v[13] * s7;
124: x[oidx + 2] -= v[14] * s1 + v[15] * s2 + v[16] * s3 + v[17] * s4 + v[18] * s5 + v[19] * s6 + v[20] * s7;
125: x[oidx + 3] -= v[21] * s1 + v[22] * s2 + v[23] * s3 + v[24] * s4 + v[25] * s5 + v[26] * s6 + v[27] * s7;
126: x[oidx + 4] -= v[28] * s1 + v[29] * s2 + v[30] * s3 + v[31] * s4 + v[32] * s5 + v[33] * s6 + v[34] * s7;
127: x[oidx + 5] -= v[35] * s1 + v[36] * s2 + v[37] * s3 + v[38] * s4 + v[39] * s5 + v[40] * s6 + v[41] * s7;
128: x[oidx + 6] -= v[42] * s1 + v[43] * s2 + v[44] * s3 + v[45] * s4 + v[46] * s5 + v[47] * s6 + v[48] * s7;
129: v -= bs2;
130: }
131: x[idx] = s1;
132: x[1 + idx] = s2;
133: x[2 + idx] = s3;
134: x[3 + idx] = s4;
135: x[4 + idx] = s5;
136: x[5 + idx] = s6;
137: x[6 + idx] = s7;
138: idx += bs;
139: }
140: /* backward solve the L^T */
141: for (i = n - 1; i >= 0; i--) {
142: v = aa + bs2 * ai[i];
143: vi = aj + ai[i];
144: nz = ai[i + 1] - ai[i];
145: idt = bs * i;
146: s1 = x[idt];
147: s2 = x[1 + idt];
148: s3 = x[2 + idt];
149: s4 = x[3 + idt];
150: s5 = x[4 + idt];
151: s6 = x[5 + idt];
152: s7 = x[6 + idt];
153: for (j = 0; j < nz; j++) {
154: idx = bs * vi[j];
155: x[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5 + v[5] * s6 + v[6] * s7;
156: x[idx + 1] -= v[7] * s1 + v[8] * s2 + v[9] * s3 + v[10] * s4 + v[11] * s5 + v[12] * s6 + v[13] * s7;
157: x[idx + 2] -= v[14] * s1 + v[15] * s2 + v[16] * s3 + v[17] * s4 + v[18] * s5 + v[19] * s6 + v[20] * s7;
158: x[idx + 3] -= v[21] * s1 + v[22] * s2 + v[23] * s3 + v[24] * s4 + v[25] * s5 + v[26] * s6 + v[27] * s7;
159: x[idx + 4] -= v[28] * s1 + v[29] * s2 + v[30] * s3 + v[31] * s4 + v[32] * s5 + v[33] * s6 + v[34] * s7;
160: x[idx + 5] -= v[35] * s1 + v[36] * s2 + v[37] * s3 + v[38] * s4 + v[39] * s5 + v[40] * s6 + v[41] * s7;
161: x[idx + 6] -= v[42] * s1 + v[43] * s2 + v[44] * s3 + v[45] * s4 + v[46] * s5 + v[47] * s6 + v[48] * s7;
162: v += bs2;
163: }
164: }
165: VecRestoreArray(xx, &x);
166: PetscLogFlops(2.0 * bs2 * (a->nz) - bs * A->cmap->n);
167: return 0;
168: }