Actual source code: baijsolvnat6.c
petsc-3.10.5 2019-03-28
1: #include <../src/mat/impls/baij/seq/baij.h>
2: #include <petsc/private/kernels/blockinvert.h>
5: PetscErrorCode MatSolve_SeqBAIJ_6_NaturalOrdering_inplace(Mat A,Vec bb,Vec xx)
6: {
7: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
8: PetscInt i,nz,idx,idt,jdx;
9: PetscErrorCode ierr;
10: const PetscInt *diag = a->diag,*vi,n=a->mbs,*ai=a->i,*aj=a->j;
11: const MatScalar *aa =a->a,*v;
12: PetscScalar *x,s1,s2,s3,s4,s5,s6,x1,x2,x3,x4,x5,x6;
13: const PetscScalar *b;
16: VecGetArrayRead(bb,&b);
17: VecGetArray(xx,&x);
18: /* forward solve the lower triangular */
19: idx = 0;
20: x[0] = b[idx]; x[1] = b[1+idx]; x[2] = b[2+idx];
21: x[3] = b[3+idx]; x[4] = b[4+idx]; x[5] = b[5+idx];
22: for (i=1; i<n; i++) {
23: v = aa + 36*ai[i];
24: vi = aj + ai[i];
25: nz = diag[i] - ai[i];
26: idx = 6*i;
27: s1 = b[idx]; s2 = b[1+idx]; s3 = b[2+idx];
28: s4 = b[3+idx]; s5 = b[4+idx]; s6 = b[5+idx];
29: while (nz--) {
30: jdx = 6*(*vi++);
31: x1 = x[jdx]; x2 = x[1+jdx]; x3 = x[2+jdx];
32: x4 = x[3+jdx]; x5 = x[4+jdx]; x6 = x[5+jdx];
33: s1 -= v[0]*x1 + v[6]*x2 + v[12]*x3 + v[18]*x4 + v[24]*x5 + v[30]*x6;
34: s2 -= v[1]*x1 + v[7]*x2 + v[13]*x3 + v[19]*x4 + v[25]*x5 + v[31]*x6;
35: s3 -= v[2]*x1 + v[8]*x2 + v[14]*x3 + v[20]*x4 + v[26]*x5 + v[32]*x6;
36: s4 -= v[3]*x1 + v[9]*x2 + v[15]*x3 + v[21]*x4 + v[27]*x5 + v[33]*x6;
37: s5 -= v[4]*x1 + v[10]*x2 + v[16]*x3 + v[22]*x4 + v[28]*x5 + v[34]*x6;
38: s6 -= v[5]*x1 + v[11]*x2 + v[17]*x3 + v[23]*x4 + v[29]*x5 + v[35]*x6;
39: v += 36;
40: }
41: x[idx] = s1;
42: x[1+idx] = s2;
43: x[2+idx] = s3;
44: x[3+idx] = s4;
45: x[4+idx] = s5;
46: x[5+idx] = s6;
47: }
48: /* backward solve the upper triangular */
49: for (i=n-1; i>=0; i--) {
50: v = aa + 36*diag[i] + 36;
51: vi = aj + diag[i] + 1;
52: nz = ai[i+1] - diag[i] - 1;
53: idt = 6*i;
54: s1 = x[idt]; s2 = x[1+idt];
55: s3 = x[2+idt]; s4 = x[3+idt];
56: s5 = x[4+idt]; s6 = x[5+idt];
57: while (nz--) {
58: idx = 6*(*vi++);
59: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx];
60: x4 = x[3+idx]; x5 = x[4+idx]; x6 = x[5+idx];
61: s1 -= v[0]*x1 + v[6]*x2 + v[12]*x3 + v[18]*x4 + v[24]*x5 + v[30]*x6;
62: s2 -= v[1]*x1 + v[7]*x2 + v[13]*x3 + v[19]*x4 + v[25]*x5 + v[31]*x6;
63: s3 -= v[2]*x1 + v[8]*x2 + v[14]*x3 + v[20]*x4 + v[26]*x5 + v[32]*x6;
64: s4 -= v[3]*x1 + v[9]*x2 + v[15]*x3 + v[21]*x4 + v[27]*x5 + v[33]*x6;
65: s5 -= v[4]*x1 + v[10]*x2 + v[16]*x3 + v[22]*x4 + v[28]*x5 + v[34]*x6;
66: s6 -= v[5]*x1 + v[11]*x2 + v[17]*x3 + v[23]*x4 + v[29]*x5 + v[35]*x6;
67: v += 36;
68: }
69: v = aa + 36*diag[i];
70: x[idt] = v[0]*s1 + v[6]*s2 + v[12]*s3 + v[18]*s4 + v[24]*s5 + v[30]*s6;
71: x[1+idt] = v[1]*s1 + v[7]*s2 + v[13]*s3 + v[19]*s4 + v[25]*s5 + v[31]*s6;
72: x[2+idt] = v[2]*s1 + v[8]*s2 + v[14]*s3 + v[20]*s4 + v[26]*s5 + v[32]*s6;
73: x[3+idt] = v[3]*s1 + v[9]*s2 + v[15]*s3 + v[21]*s4 + v[27]*s5 + v[33]*s6;
74: x[4+idt] = v[4]*s1 + v[10]*s2 + v[16]*s3 + v[22]*s4 + v[28]*s5 + v[34]*s6;
75: x[5+idt] = v[5]*s1 + v[11]*s2 + v[17]*s3 + v[23]*s4 + v[29]*s5 + v[35]*s6;
76: }
78: VecRestoreArrayRead(bb,&b);
79: VecRestoreArray(xx,&x);
80: PetscLogFlops(2.0*36*(a->nz) - 6.0*A->cmap->n);
81: return(0);
82: }
84: PetscErrorCode MatSolve_SeqBAIJ_6_NaturalOrdering(Mat A,Vec bb,Vec xx)
85: {
86: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
87: const PetscInt n =a->mbs,*vi,*ai=a->i,*aj=a->j,*adiag=a->diag;
88: PetscErrorCode ierr;
89: PetscInt i,k,nz,idx,jdx,idt;
90: const PetscInt bs = A->rmap->bs,bs2 = a->bs2;
91: const MatScalar *aa=a->a,*v;
92: PetscScalar *x;
93: const PetscScalar *b;
94: PetscScalar s1,s2,s3,s4,s5,s6,x1,x2,x3,x4,x5,x6;
97: VecGetArrayRead(bb,&b);
98: VecGetArray(xx,&x);
99: /* forward solve the lower triangular */
100: idx = 0;
101: x[0] = b[idx]; x[1] = b[1+idx];x[2] = b[2+idx];x[3] = b[3+idx];
102: x[4] = b[4+idx];x[5] = b[5+idx];
103: for (i=1; i<n; i++) {
104: v = aa + bs2*ai[i];
105: vi = aj + ai[i];
106: nz = ai[i+1] - ai[i];
107: idx = bs*i;
108: s1 = b[idx];s2 = b[1+idx];s3 = b[2+idx];s4 = b[3+idx];
109: s5 = b[4+idx];s6 = b[5+idx];
110: for (k=0; k<nz; k++) {
111: jdx = bs*vi[k];
112: x1 = x[jdx];x2 = x[1+jdx]; x3 =x[2+jdx];x4 =x[3+jdx];
113: x5 = x[4+jdx]; x6 = x[5+jdx];
114: s1 -= v[0]*x1 + v[6]*x2 + v[12]*x3 + v[18]*x4 + v[24]*x5 + v[30]*x6;
115: s2 -= v[1]*x1 + v[7]*x2 + v[13]*x3 + v[19]*x4 + v[25]*x5 + v[31]*x6;;
116: s3 -= v[2]*x1 + v[8]*x2 + v[14]*x3 + v[20]*x4 + v[26]*x5 + v[32]*x6;
117: s4 -= v[3]*x1 + v[9]*x2 + v[15]*x3 + v[21]*x4 + v[27]*x5 + v[33]*x6;
118: s5 -= v[4]*x1 + v[10]*x2 + v[16]*x3 + v[22]*x4 + v[28]*x5 + v[34]*x6;
119: s6 -= v[5]*x1 + v[11]*x2 + v[17]*x3 + v[23]*x4 + v[29]*x5 + v[35]*x6;
120: v += bs2;
121: }
123: x[idx] = s1;
124: x[1+idx] = s2;
125: x[2+idx] = s3;
126: x[3+idx] = s4;
127: x[4+idx] = s5;
128: x[5+idx] = s6;
129: }
131: /* backward solve the upper triangular */
132: for (i=n-1; i>=0; i--) {
133: v = aa + bs2*(adiag[i+1]+1);
134: vi = aj + adiag[i+1]+1;
135: nz = adiag[i] - adiag[i+1]-1;
136: idt = bs*i;
137: s1 = x[idt]; s2 = x[1+idt];s3 = x[2+idt];s4 = x[3+idt];
138: s5 = x[4+idt];s6 = x[5+idt];
139: for (k=0; k<nz; k++) {
140: idx = bs*vi[k];
141: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx];x4 = x[3+idx];
142: x5 = x[4+idx];x6 = x[5+idx];
143: s1 -= v[0]*x1 + v[6]*x2 + v[12]*x3 + v[18]*x4 + v[24]*x5 + v[30]*x6;
144: s2 -= v[1]*x1 + v[7]*x2 + v[13]*x3 + v[19]*x4 + v[25]*x5 + v[31]*x6;;
145: s3 -= v[2]*x1 + v[8]*x2 + v[14]*x3 + v[20]*x4 + v[26]*x5 + v[32]*x6;
146: s4 -= v[3]*x1 + v[9]*x2 + v[15]*x3 + v[21]*x4 + v[27]*x5 + v[33]*x6;
147: s5 -= v[4]*x1 + v[10]*x2 + v[16]*x3 + v[22]*x4 + v[28]*x5 + v[34]*x6;
148: s6 -= v[5]*x1 + v[11]*x2 + v[17]*x3 + v[23]*x4 + v[29]*x5 + v[35]*x6;
149: v += bs2;
150: }
151: /* x = inv_diagonal*x */
152: x[idt] = v[0]*s1 + v[6]*s2 + v[12]*s3 + v[18]*s4 + v[24]*s5 + v[30]*s6;
153: x[1+idt] = v[1]*s1 + v[7]*s2 + v[13]*s3 + v[19]*s4 + v[25]*s5 + v[31]*s6;
154: x[2+idt] = v[2]*s1 + v[8]*s2 + v[14]*s3 + v[20]*s4 + v[26]*s5 + v[32]*s6;
155: x[3+idt] = v[3]*s1 + v[9]*s2 + v[15]*s3 + v[21]*s4 + v[27]*s5 + v[33]*s6;
156: x[4+idt] = v[4]*s1 + v[10]*s2 + v[16]*s3 + v[22]*s4 + v[28]*s5 + v[34]*s6;
157: x[5+idt] = v[5]*s1 + v[11]*s2 + v[17]*s3 + v[23]*s4 + v[29]*s5 + v[35]*s6;
158: }
160: VecRestoreArrayRead(bb,&b);
161: VecRestoreArray(xx,&x);
162: PetscLogFlops(2.0*bs2*(a->nz) - bs*A->cmap->n);
163: return(0);
164: }