Actual source code: dgefa5.c
1: #define PETSCMAT_DLL
3: /*
4: Inverts 5 by 5 matrix using partial pivoting.
6: Used by the sparse factorization routines in
7: src/mat/impls/baij/seq
9: This is a combination of the Linpack routines
10: dgefa() and dgedi() specialized for a size of 5.
12: */
13: #include petscsys.h
17: PetscErrorCode Kernel_A_gets_inverse_A_5(MatScalar *a,PetscInt *ipvt,MatScalar *work,PetscReal shift)
18: {
19: PetscInt i__2,i__3,kp1,j,k,l,ll,i,kb,k3;
20: PetscInt k4,j3;
21: MatScalar *aa,*ax,*ay,stmp;
22: MatReal tmp,max;
24: /* gaussian elimination with partial pivoting */
27: /* Parameter adjustments */
28: a -= 6;
30: for (k = 1; k <= 4; ++k) {
31: kp1 = k + 1;
32: k3 = 5*k;
33: k4 = k3 + k;
34: /* find l = pivot index */
36: i__2 = 6 - k;
37: aa = &a[k4];
38: max = PetscAbsScalar(aa[0]);
39: l = 1;
40: for (ll=1; ll<i__2; ll++) {
41: tmp = PetscAbsScalar(aa[ll]);
42: if (tmp > max) { max = tmp; l = ll+1;}
43: }
44: l += k - 1;
45: ipvt[k-1] = l;
47: if (a[l + k3] == 0.0) {
48: SETERRQ1(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot, row %D",k-1);
49: }
51: /* interchange if necessary */
53: if (l != k) {
54: stmp = a[l + k3];
55: a[l + k3] = a[k4];
56: a[k4] = stmp;
57: }
59: /* compute multipliers */
61: stmp = -1. / a[k4];
62: i__2 = 5 - k;
63: aa = &a[1 + k4];
64: for (ll=0; ll<i__2; ll++) {
65: aa[ll] *= stmp;
66: }
68: /* row elimination with column indexing */
70: ax = &a[k4+1];
71: for (j = kp1; j <= 5; ++j) {
72: j3 = 5*j;
73: stmp = a[l + j3];
74: if (l != k) {
75: a[l + j3] = a[k + j3];
76: a[k + j3] = stmp;
77: }
79: i__3 = 5 - k;
80: ay = &a[1+k+j3];
81: for (ll=0; ll<i__3; ll++) {
82: ay[ll] += stmp*ax[ll];
83: }
84: }
85: }
86: ipvt[4] = 5;
87: if (a[30] == 0.0) {
88: SETERRQ1(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot, row %D",4);
89: }
91: /*
92: Now form the inverse
93: */
95: /* compute inverse(u) */
97: for (k = 1; k <= 5; ++k) {
98: k3 = 5*k;
99: k4 = k3 + k;
100: a[k4] = 1.0 / a[k4];
101: stmp = -a[k4];
102: i__2 = k - 1;
103: aa = &a[k3 + 1];
104: for (ll=0; ll<i__2; ll++) aa[ll] *= stmp;
105: kp1 = k + 1;
106: if (5 < kp1) continue;
107: ax = aa;
108: for (j = kp1; j <= 5; ++j) {
109: j3 = 5*j;
110: stmp = a[k + j3];
111: a[k + j3] = 0.0;
112: ay = &a[j3 + 1];
113: for (ll=0; ll<k; ll++) {
114: ay[ll] += stmp*ax[ll];
115: }
116: }
117: }
119: /* form inverse(u)*inverse(l) */
121: for (kb = 1; kb <= 4; ++kb) {
122: k = 5 - kb;
123: k3 = 5*k;
124: kp1 = k + 1;
125: aa = a + k3;
126: for (i = kp1; i <= 5; ++i) {
127: work[i-1] = aa[i];
128: aa[i] = 0.0;
129: }
130: for (j = kp1; j <= 5; ++j) {
131: stmp = work[j-1];
132: ax = &a[5*j + 1];
133: ay = &a[k3 + 1];
134: ay[0] += stmp*ax[0];
135: ay[1] += stmp*ax[1];
136: ay[2] += stmp*ax[2];
137: ay[3] += stmp*ax[3];
138: ay[4] += stmp*ax[4];
139: }
140: l = ipvt[k-1];
141: if (l != k) {
142: ax = &a[k3 + 1];
143: ay = &a[5*l + 1];
144: stmp = ax[0]; ax[0] = ay[0]; ay[0] = stmp;
145: stmp = ax[1]; ax[1] = ay[1]; ay[1] = stmp;
146: stmp = ax[2]; ax[2] = ay[2]; ay[2] = stmp;
147: stmp = ax[3]; ax[3] = ay[3]; ay[3] = stmp;
148: stmp = ax[4]; ax[4] = ay[4]; ay[4] = stmp;
149: }
150: }
151: return(0);
152: }