Actual source code: baijsolvtran5.c
1: #include <../src/mat/impls/baij/seq/baij.h>
2: #include <petsc/private/kernels/blockinvert.h>
4: PetscErrorCode MatSolveTranspose_SeqBAIJ_5_inplace(Mat A, Vec bb, Vec xx)
5: {
6: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
7: IS iscol = a->col, isrow = a->row;
8: const PetscInt *r, *c, *rout, *cout;
9: const PetscInt *diag = a->diag, n = a->mbs, *vi, *ai = a->i, *aj = a->j;
10: PetscInt i, nz, idx, idt, ii, ic, ir, oidx;
11: const MatScalar *aa = a->a, *v;
12: PetscScalar s1, s2, s3, s4, s5, x1, x2, x3, x4, x5, *x, *t;
13: const PetscScalar *b;
15: PetscFunctionBegin;
16: PetscCall(VecGetArrayRead(bb, &b));
17: PetscCall(VecGetArray(xx, &x));
18: t = a->solve_work;
20: PetscCall(ISGetIndices(isrow, &rout));
21: r = rout;
22: PetscCall(ISGetIndices(iscol, &cout));
23: c = cout;
25: /* copy the b into temp work space according to permutation */
26: ii = 0;
27: for (i = 0; i < n; i++) {
28: ic = 5 * c[i];
29: t[ii] = b[ic];
30: t[ii + 1] = b[ic + 1];
31: t[ii + 2] = b[ic + 2];
32: t[ii + 3] = b[ic + 3];
33: t[ii + 4] = b[ic + 4];
34: ii += 5;
35: }
37: /* forward solve the U^T */
38: idx = 0;
39: for (i = 0; i < n; i++) {
40: v = aa + 25 * diag[i];
41: /* multiply by the inverse of the block diagonal */
42: x1 = t[idx];
43: x2 = t[1 + idx];
44: x3 = t[2 + idx];
45: x4 = t[3 + idx];
46: x5 = t[4 + idx];
47: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3 + v[3] * x4 + v[4] * x5;
48: s2 = v[5] * x1 + v[6] * x2 + v[7] * x3 + v[8] * x4 + v[9] * x5;
49: s3 = v[10] * x1 + v[11] * x2 + v[12] * x3 + v[13] * x4 + v[14] * x5;
50: s4 = v[15] * x1 + v[16] * x2 + v[17] * x3 + v[18] * x4 + v[19] * x5;
51: s5 = v[20] * x1 + v[21] * x2 + v[22] * x3 + v[23] * x4 + v[24] * x5;
52: v += 25;
54: vi = aj + diag[i] + 1;
55: nz = ai[i + 1] - diag[i] - 1;
56: while (nz--) {
57: oidx = 5 * (*vi++);
58: t[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5;
59: t[oidx + 1] -= v[5] * s1 + v[6] * s2 + v[7] * s3 + v[8] * s4 + v[9] * s5;
60: t[oidx + 2] -= v[10] * s1 + v[11] * s2 + v[12] * s3 + v[13] * s4 + v[14] * s5;
61: t[oidx + 3] -= v[15] * s1 + v[16] * s2 + v[17] * s3 + v[18] * s4 + v[19] * s5;
62: t[oidx + 4] -= v[20] * s1 + v[21] * s2 + v[22] * s3 + v[23] * s4 + v[24] * s5;
63: v += 25;
64: }
65: t[idx] = s1;
66: t[1 + idx] = s2;
67: t[2 + idx] = s3;
68: t[3 + idx] = s4;
69: t[4 + idx] = s5;
70: idx += 5;
71: }
72: /* backward solve the L^T */
73: for (i = n - 1; i >= 0; i--) {
74: v = aa + 25 * diag[i] - 25;
75: vi = aj + diag[i] - 1;
76: nz = diag[i] - ai[i];
77: idt = 5 * i;
78: s1 = t[idt];
79: s2 = t[1 + idt];
80: s3 = t[2 + idt];
81: s4 = t[3 + idt];
82: s5 = t[4 + idt];
83: while (nz--) {
84: idx = 5 * (*vi--);
85: t[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5;
86: t[idx + 1] -= v[5] * s1 + v[6] * s2 + v[7] * s3 + v[8] * s4 + v[9] * s5;
87: t[idx + 2] -= v[10] * s1 + v[11] * s2 + v[12] * s3 + v[13] * s4 + v[14] * s5;
88: t[idx + 3] -= v[15] * s1 + v[16] * s2 + v[17] * s3 + v[18] * s4 + v[19] * s5;
89: t[idx + 4] -= v[20] * s1 + v[21] * s2 + v[22] * s3 + v[23] * s4 + v[24] * s5;
90: v -= 25;
91: }
92: }
94: /* copy t into x according to permutation */
95: ii = 0;
96: for (i = 0; i < n; i++) {
97: ir = 5 * r[i];
98: x[ir] = t[ii];
99: x[ir + 1] = t[ii + 1];
100: x[ir + 2] = t[ii + 2];
101: x[ir + 3] = t[ii + 3];
102: x[ir + 4] = t[ii + 4];
103: ii += 5;
104: }
106: PetscCall(ISRestoreIndices(isrow, &rout));
107: PetscCall(ISRestoreIndices(iscol, &cout));
108: PetscCall(VecRestoreArrayRead(bb, &b));
109: PetscCall(VecRestoreArray(xx, &x));
110: PetscCall(PetscLogFlops(2.0 * 25 * (a->nz) - 5.0 * A->cmap->n));
111: PetscFunctionReturn(PETSC_SUCCESS);
112: }
114: PetscErrorCode MatSolveTranspose_SeqBAIJ_5(Mat A, Vec bb, Vec xx)
115: {
116: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
117: IS iscol = a->col, isrow = a->row;
118: const PetscInt n = a->mbs, *vi, *ai = a->i, *aj = a->j, *diag = a->diag;
119: const PetscInt *r, *c, *rout, *cout;
120: PetscInt nz, idx, idt, j, i, oidx, ii, ic, ir;
121: const PetscInt bs = A->rmap->bs, bs2 = a->bs2;
122: const MatScalar *aa = a->a, *v;
123: PetscScalar s1, s2, s3, s4, s5, x1, x2, x3, x4, x5, *x, *t;
124: const PetscScalar *b;
126: PetscFunctionBegin;
127: PetscCall(VecGetArrayRead(bb, &b));
128: PetscCall(VecGetArray(xx, &x));
129: t = a->solve_work;
131: PetscCall(ISGetIndices(isrow, &rout));
132: r = rout;
133: PetscCall(ISGetIndices(iscol, &cout));
134: c = cout;
136: /* copy b into temp work space according to permutation */
137: for (i = 0; i < n; i++) {
138: ii = bs * i;
139: ic = bs * c[i];
140: t[ii] = b[ic];
141: t[ii + 1] = b[ic + 1];
142: t[ii + 2] = b[ic + 2];
143: t[ii + 3] = b[ic + 3];
144: t[ii + 4] = b[ic + 4];
145: }
147: /* forward solve the U^T */
148: idx = 0;
149: for (i = 0; i < n; i++) {
150: v = aa + bs2 * diag[i];
151: /* multiply by the inverse of the block diagonal */
152: x1 = t[idx];
153: x2 = t[1 + idx];
154: x3 = t[2 + idx];
155: x4 = t[3 + idx];
156: x5 = t[4 + idx];
157: s1 = v[0] * x1 + v[1] * x2 + v[2] * x3 + v[3] * x4 + v[4] * x5;
158: s2 = v[5] * x1 + v[6] * x2 + v[7] * x3 + v[8] * x4 + v[9] * x5;
159: s3 = v[10] * x1 + v[11] * x2 + v[12] * x3 + v[13] * x4 + v[14] * x5;
160: s4 = v[15] * x1 + v[16] * x2 + v[17] * x3 + v[18] * x4 + v[19] * x5;
161: s5 = v[20] * x1 + v[21] * x2 + v[22] * x3 + v[23] * x4 + v[24] * x5;
162: v -= bs2;
164: vi = aj + diag[i] - 1;
165: nz = diag[i] - diag[i + 1] - 1;
166: for (j = 0; j > -nz; j--) {
167: oidx = bs * vi[j];
168: t[oidx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5;
169: t[oidx + 1] -= v[5] * s1 + v[6] * s2 + v[7] * s3 + v[8] * s4 + v[9] * s5;
170: t[oidx + 2] -= v[10] * s1 + v[11] * s2 + v[12] * s3 + v[13] * s4 + v[14] * s5;
171: t[oidx + 3] -= v[15] * s1 + v[16] * s2 + v[17] * s3 + v[18] * s4 + v[19] * s5;
172: t[oidx + 4] -= v[20] * s1 + v[21] * s2 + v[22] * s3 + v[23] * s4 + v[24] * s5;
173: v -= bs2;
174: }
175: t[idx] = s1;
176: t[1 + idx] = s2;
177: t[2 + idx] = s3;
178: t[3 + idx] = s4;
179: t[4 + idx] = s5;
180: idx += bs;
181: }
182: /* backward solve the L^T */
183: for (i = n - 1; i >= 0; i--) {
184: v = aa + bs2 * ai[i];
185: vi = aj + ai[i];
186: nz = ai[i + 1] - ai[i];
187: idt = bs * i;
188: s1 = t[idt];
189: s2 = t[1 + idt];
190: s3 = t[2 + idt];
191: s4 = t[3 + idt];
192: s5 = t[4 + idt];
193: for (j = 0; j < nz; j++) {
194: idx = bs * vi[j];
195: t[idx] -= v[0] * s1 + v[1] * s2 + v[2] * s3 + v[3] * s4 + v[4] * s5;
196: t[idx + 1] -= v[5] * s1 + v[6] * s2 + v[7] * s3 + v[8] * s4 + v[9] * s5;
197: t[idx + 2] -= v[10] * s1 + v[11] * s2 + v[12] * s3 + v[13] * s4 + v[14] * s5;
198: t[idx + 3] -= v[15] * s1 + v[16] * s2 + v[17] * s3 + v[18] * s4 + v[19] * s5;
199: t[idx + 4] -= v[20] * s1 + v[21] * s2 + v[22] * s3 + v[23] * s4 + v[24] * s5;
200: v += bs2;
201: }
202: }
204: /* copy t into x according to permutation */
205: for (i = 0; i < n; i++) {
206: ii = bs * i;
207: ir = bs * r[i];
208: x[ir] = t[ii];
209: x[ir + 1] = t[ii + 1];
210: x[ir + 2] = t[ii + 2];
211: x[ir + 3] = t[ii + 3];
212: x[ir + 4] = t[ii + 4];
213: }
215: PetscCall(ISRestoreIndices(isrow, &rout));
216: PetscCall(ISRestoreIndices(iscol, &cout));
217: PetscCall(VecRestoreArrayRead(bb, &b));
218: PetscCall(VecRestoreArray(xx, &x));
219: PetscCall(PetscLogFlops(2.0 * bs2 * (a->nz) - bs * A->cmap->n));
220: PetscFunctionReturn(PETSC_SUCCESS);
221: }