Actual source code: baijsolvtrannat7.c
petsc-3.12.5 2020-03-29
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: PetscErrorCode ierr;
7: const PetscInt *diag=a->diag,n=a->mbs,*vi,*ai=a->i,*aj=a->j;
8: PetscInt i,nz,idx,idt,oidx;
9: const MatScalar *aa=a->a,*v;
10: PetscScalar s1,s2,s3,s4,s5,s6,s7,x1,x2,x3,x4,x5,x6,x7,*x;
13: VecCopy(bb,xx);
14: VecGetArray(xx,&x);
16: /* forward solve the U^T */
17: idx = 0;
18: for (i=0; i<n; i++) {
20: v = aa + 49*diag[i];
21: /* multiply by the inverse of the block diagonal */
22: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx]; x4 = x[3+idx]; x5 = x[4+idx];
23: x6 = x[5+idx]; x7 = x[6+idx];
24: s1 = v[0]*x1 + v[1]*x2 + v[2]*x3 + v[3]*x4 + v[4]*x5 + v[5]*x6 + v[6]*x7;
25: s2 = v[7]*x1 + v[8]*x2 + v[9]*x3 + v[10]*x4 + v[11]*x5 + v[12]*x6 + v[13]*x7;
26: s3 = v[14]*x1 + v[15]*x2 + v[16]*x3 + v[17]*x4 + v[18]*x5 + v[19]*x6 + v[20]*x7;
27: s4 = v[21]*x1 + v[22]*x2 + v[23]*x3 + v[24]*x4 + v[25]*x5 + v[26]*x6 + v[27]*x7;
28: s5 = v[28]*x1 + v[29]*x2 + v[30]*x3 + v[31]*x4 + v[32]*x5 + v[33]*x6 + v[34]*x7;
29: s6 = v[35]*x1 + v[36]*x2 + v[37]*x3 + v[38]*x4 + v[39]*x5 + v[40]*x6 + v[41]*x7;
30: s7 = v[42]*x1 + v[43]*x2 + v[44]*x3 + v[45]*x4 + v[46]*x5 + v[47]*x6 + v[48]*x7;
31: v += 49;
33: vi = aj + diag[i] + 1;
34: nz = ai[i+1] - diag[i] - 1;
35: while (nz--) {
36: oidx = 7*(*vi++);
37: x[oidx] -= v[0]*s1 + v[1]*s2 + v[2]*s3 + v[3]*s4 + v[4]*s5 + v[5]*s6 + v[6]*s7;
38: x[oidx+1] -= v[7]*s1 + v[8]*s2 + v[9]*s3 + v[10]*s4 + v[11]*s5 + v[12]*s6 + v[13]*s7;
39: x[oidx+2] -= v[14]*s1 + v[15]*s2 + v[16]*s3 + v[17]*s4 + v[18]*s5 + v[19]*s6 + v[20]*s7;
40: x[oidx+3] -= v[21]*s1 + v[22]*s2 + v[23]*s3 + v[24]*s4 + v[25]*s5 + v[26]*s6 + v[27]*s7;
41: x[oidx+4] -= v[28]*s1 + v[29]*s2 + v[30]*s3 + v[31]*s4 + v[32]*s5 + v[33]*s6 + v[34]*s7;
42: x[oidx+5] -= v[35]*s1 + v[36]*s2 + v[37]*s3 + v[38]*s4 + v[39]*s5 + v[40]*s6 + v[41]*s7;
43: x[oidx+6] -= v[42]*s1 + v[43]*s2 + v[44]*s3 + v[45]*s4 + v[46]*s5 + v[47]*s6 + v[48]*s7;
44: v += 49;
45: }
46: x[idx] = s1;x[1+idx] = s2; x[2+idx] = s3;x[3+idx] = s4; x[4+idx] = s5;
47: x[5+idx] = s6;x[6+idx] = s7;
48: idx += 7;
49: }
50: /* backward solve the L^T */
51: for (i=n-1; i>=0; i--) {
52: v = aa + 49*diag[i] - 49;
53: vi = aj + diag[i] - 1;
54: nz = diag[i] - ai[i];
55: idt = 7*i;
56: s1 = x[idt]; s2 = x[1+idt]; s3 = x[2+idt];s4 = x[3+idt]; s5 = x[4+idt];
57: s6 = x[5+idt];s7 = x[6+idt];
58: while (nz--) {
59: idx = 7*(*vi--);
60: x[idx] -= v[0]*s1 + v[1]*s2 + v[2]*s3 + v[3]*s4 + v[4]*s5 + v[5]*s6 + v[6]*s7;
61: x[idx+1] -= v[7]*s1 + v[8]*s2 + v[9]*s3 + v[10]*s4 + v[11]*s5 + v[12]*s6 + v[13]*s7;
62: x[idx+2] -= v[14]*s1 + v[15]*s2 + v[16]*s3 + v[17]*s4 + v[18]*s5 + v[19]*s6 + v[20]*s7;
63: x[idx+3] -= v[21]*s1 + v[22]*s2 + v[23]*s3 + v[24]*s4 + v[25]*s5 + v[26]*s6 + v[27]*s7;
64: x[idx+4] -= v[28]*s1 + v[29]*s2 + v[30]*s3 + v[31]*s4 + v[32]*s5 + v[33]*s6 + v[34]*s7;
65: x[idx+5] -= v[35]*s1 + v[36]*s2 + v[37]*s3 + v[38]*s4 + v[39]*s5 + v[40]*s6 + v[41]*s7;
66: x[idx+6] -= v[42]*s1 + v[43]*s2 + v[44]*s3 + v[45]*s4 + v[46]*s5 + v[47]*s6 + v[48]*s7;
67: v -= 49;
68: }
69: }
70: VecRestoreArray(xx,&x);
71: PetscLogFlops(2.0*49*(a->nz) - 7.0*A->cmap->n);
72: return(0);
73: }
74: PetscErrorCode MatSolveTranspose_SeqBAIJ_7_NaturalOrdering(Mat A,Vec bb,Vec xx)
75: {
76: Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data;
77: PetscErrorCode ierr;
78: const PetscInt n=a->mbs,*vi,*ai=a->i,*aj=a->j,*diag=a->diag;
79: PetscInt nz,idx,idt,j,i,oidx;
80: const PetscInt bs =A->rmap->bs,bs2=a->bs2;
81: const MatScalar *aa=a->a,*v;
82: PetscScalar s1,s2,s3,s4,s5,s6,s7,x1,x2,x3,x4,x5,x6,x7,*x;
85: VecCopy(bb,xx);
86: VecGetArray(xx,&x);
88: /* forward solve the U^T */
89: idx = 0;
90: for (i=0; i<n; i++) {
91: v = aa + bs2*diag[i];
92: /* multiply by the inverse of the block diagonal */
93: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx]; x4 = x[3+idx];
94: x5 = x[4+idx]; x6 = x[5+idx]; x7 = x[6+idx];
95: s1 = v[0]*x1 + v[1]*x2 + v[2]*x3 + v[3]*x4 + v[4]*x5 + v[5]*x6 + v[6]*x7;
96: s2 = v[7]*x1 + v[8]*x2 + v[9]*x3 + v[10]*x4 + v[11]*x5 + v[12]*x6 + v[13]*x7;
97: s3 = v[14]*x1 + v[15]*x2 + v[16]*x3 + v[17]*x4 + v[18]*x5 + v[19]*x6 + v[20]*x7;
98: s4 = v[21]*x1 + v[22]*x2 + v[23]*x3 + v[24]*x4 + v[25]*x5 + v[26]*x6 + v[27]*x7;
99: s5 = v[28]*x1 + v[29]*x2 + v[30]*x3 + v[31]*x4 + v[32]*x5 + v[33]*x6 + v[34]*x7;
100: s6 = v[35]*x1 + v[36]*x2 + v[37]*x3 + v[38]*x4 + v[39]*x5 + v[40]*x6 + v[41]*x7;
101: s7 = v[42]*x1 + v[43]*x2 + v[44]*x3 + v[45]*x4 + v[46]*x5 + v[47]*x6 + v[48]*x7;
102: v -= bs2;
103: vi = aj + diag[i] - 1;
104: nz = diag[i] - diag[i+1] - 1;
105: for (j=0; j>-nz; j--) {
106: oidx = bs*vi[j];
107: x[oidx] -= v[0]*s1 + v[1]*s2 + v[2]*s3 + v[3]*s4 + v[4]*s5 + v[5]*s6 + v[6]*s7;
108: x[oidx+1] -= v[7]*s1 + v[8]*s2 + v[9]*s3 + v[10]*s4 + v[11]*s5 + v[12]*s6 + v[13]*s7;
109: x[oidx+2] -= v[14]*s1 + v[15]*s2 + v[16]*s3 + v[17]*s4 + v[18]*s5 + v[19]*s6 + v[20]*s7;
110: x[oidx+3] -= v[21]*s1 + v[22]*s2 + v[23]*s3 + v[24]*s4 + v[25]*s5 + v[26]*s6 + v[27]*s7;
111: x[oidx+4] -= v[28]*s1 + v[29]*s2 + v[30]*s3 + v[31]*s4 + v[32]*s5 + v[33]*s6 + v[34]*s7;
112: x[oidx+5] -= v[35]*s1 + v[36]*s2 + v[37]*s3 + v[38]*s4 + v[39]*s5 + v[40]*s6 + v[41]*s7;
113: x[oidx+6] -= v[42]*s1 + v[43]*s2 + v[44]*s3 + v[45]*s4 + v[46]*s5 + v[47]*s6 + v[48]*s7;
114: v -= bs2;
115: }
116: x[idx] = s1; x[1+idx] = s2; x[2+idx] = s3; x[3+idx] = s4; x[4+idx] = s5;
117: x[5+idx] = s6; x[6+idx] = s7;
118: idx += bs;
119: }
120: /* backward solve the L^T */
121: for (i=n-1; i>=0; i--) {
122: v = aa + bs2*ai[i];
123: vi = aj + ai[i];
124: nz = ai[i+1] - ai[i];
125: idt = bs*i;
126: s1 = x[idt]; s2 = x[1+idt]; s3 = x[2+idt]; s4 = x[3+idt]; s5 = x[4+idt];
127: s6 = x[5+idt]; s7 = x[6+idt];
128: for (j=0; j<nz; j++) {
129: idx = bs*vi[j];
130: x[idx] -= v[0]*s1 + v[1]*s2 + v[2]*s3 + v[3]*s4 + v[4]*s5 + v[5]*s6 + v[6]*s7;
131: x[idx+1] -= v[7]*s1 + v[8]*s2 + v[9]*s3 + v[10]*s4 + v[11]*s5 + v[12]*s6 + v[13]*s7;
132: x[idx+2] -= v[14]*s1 + v[15]*s2 + v[16]*s3 + v[17]*s4 + v[18]*s5 + v[19]*s6 + v[20]*s7;
133: x[idx+3] -= v[21]*s1 + v[22]*s2 + v[23]*s3 + v[24]*s4 + v[25]*s5 + v[26]*s6 + v[27]*s7;
134: x[idx+4] -= v[28]*s1 + v[29]*s2 + v[30]*s3 + v[31]*s4 + v[32]*s5 + v[33]*s6 + v[34]*s7;
135: x[idx+5] -= v[35]*s1 + v[36]*s2 + v[37]*s3 + v[38]*s4 + v[39]*s5 + v[40]*s6 + v[41]*s7;
136: x[idx+6] -= v[42]*s1 + v[43]*s2 + v[44]*s3 + v[45]*s4 + v[46]*s5 + v[47]*s6 + v[48]*s7;
137: v += bs2;
138: }
139: }
140: VecRestoreArray(xx,&x);
141: PetscLogFlops(2.0*bs2*(a->nz) - bs*A->cmap->n);
142: return(0);
143: }