xref: /petsc/src/mat/tests/ex28.c (revision d2dffbf663ccc3d78d082b483cae224715b9dfa9)
1 static char help[] = "Illustrate how to do one symbolic factorization and multiple numeric factorizations using same matrix nonzero structure. \n\n";
2 
3 #include <petscmat.h>
4 
5 int main(int argc, char **args)
6 {
7   PetscInt      i, rstart, rend, N = 10, num_numfac = 5, col[3], k;
8   Mat           A[5], F;
9   Vec           u, x, b;
10   PetscMPIInt   rank;
11   PetscScalar   value[3];
12   PetscReal     norm, tol = 100 * PETSC_MACHINE_EPSILON;
13   IS            perm, iperm;
14   MatFactorInfo info;
15   MatFactorType facttype = MAT_FACTOR_LU;
16   char          solvertype[64];
17   char          factortype[64];
18 
19   PetscFunctionBeginUser;
20   PetscCall(PetscInitialize(&argc, &args, NULL, help));
21   PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
22 
23   /* Create and assemble matrices, all have same data structure */
24   for (k = 0; k < num_numfac; k++) {
25     PetscCall(MatCreate(PETSC_COMM_WORLD, &A[k]));
26     PetscCall(MatSetSizes(A[k], PETSC_DECIDE, PETSC_DECIDE, N, N));
27     PetscCall(MatSetFromOptions(A[k]));
28     PetscCall(MatSetUp(A[k]));
29     PetscCall(MatGetOwnershipRange(A[k], &rstart, &rend));
30 
31     value[0] = -1.0 * (k + 1);
32     value[1] = 2.0 * (k + 1);
33     value[2] = -1.0 * (k + 1);
34     for (i = rstart; i < rend; i++) {
35       col[0] = i - 1;
36       col[1] = i;
37       col[2] = i + 1;
38       if (i == 0) {
39         PetscCall(MatSetValues(A[k], 1, &i, 2, col + 1, value + 1, INSERT_VALUES));
40       } else if (i == N - 1) {
41         PetscCall(MatSetValues(A[k], 1, &i, 2, col, value, INSERT_VALUES));
42       } else {
43         PetscCall(MatSetValues(A[k], 1, &i, 3, col, value, INSERT_VALUES));
44       }
45     }
46     PetscCall(MatAssemblyBegin(A[k], MAT_FINAL_ASSEMBLY));
47     PetscCall(MatAssemblyEnd(A[k], MAT_FINAL_ASSEMBLY));
48     PetscCall(MatSetOption(A[k], MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE));
49   }
50 
51   /* Create vectors */
52   PetscCall(MatCreateVecs(A[0], &x, &b));
53   PetscCall(VecDuplicate(x, &u));
54 
55   /* Set rhs vector b */
56   PetscCall(VecSet(b, 1.0));
57 
58   /* Get a symbolic factor F from A[0] */
59   PetscCall(PetscStrncpy(solvertype, "petsc", sizeof(solvertype)));
60   PetscCall(PetscOptionsGetString(NULL, NULL, "-mat_solver_type", solvertype, sizeof(solvertype), NULL));
61   PetscCall(PetscOptionsGetEnum(NULL, NULL, "-mat_factor_type", MatFactorTypes, (PetscEnum *)&facttype, NULL));
62 
63   PetscCall(MatGetFactor(A[0], solvertype, facttype, &F));
64   /* test mumps options */
65   PetscCall(MatMumpsSetIcntl(F, 7, 5));
66   PetscCall(PetscStrncpy(factortype, MatFactorTypes[facttype], sizeof(factortype)));
67   PetscCall(PetscStrtoupper(solvertype));
68   PetscCall(PetscStrtoupper(factortype));
69   PetscCall(PetscPrintf(PETSC_COMM_WORLD, " %s %s:\n", solvertype, factortype));
70 
71   PetscCall(MatFactorInfoInitialize(&info));
72   info.fill = 5.0;
73   PetscCall(MatGetOrdering(A[0], MATORDERINGNATURAL, &perm, &iperm));
74   switch (facttype) {
75   case MAT_FACTOR_LU:
76     PetscCall(MatLUFactorSymbolic(F, A[0], perm, iperm, &info));
77     break;
78   case MAT_FACTOR_ILU:
79     PetscCall(MatILUFactorSymbolic(F, A[0], perm, iperm, &info));
80     break;
81   case MAT_FACTOR_ICC:
82     PetscCall(MatICCFactorSymbolic(F, A[0], perm, &info));
83     break;
84   case MAT_FACTOR_CHOLESKY:
85     PetscCall(MatCholeskyFactorSymbolic(F, A[0], perm, &info));
86     break;
87   default:
88     SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_SUP, "Not for factor type %s", factortype);
89   }
90 
91   /* Compute numeric factors using same F, then solve */
92   for (k = 0; k < num_numfac; k++) {
93     switch (facttype) {
94     case MAT_FACTOR_LU:
95     case MAT_FACTOR_ILU:
96       PetscCall(MatLUFactorNumeric(F, A[k], &info));
97       break;
98     case MAT_FACTOR_ICC:
99     case MAT_FACTOR_CHOLESKY:
100       PetscCall(MatCholeskyFactorNumeric(F, A[k], &info));
101       break;
102     default:
103       SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_SUP, "Not for factor type %s", factortype);
104     }
105 
106     /* Solve A[k] * x = b */
107     PetscCall(MatSolve(F, b, x));
108 
109     /* Check the residual */
110     PetscCall(MatMult(A[k], x, u));
111     PetscCall(VecAXPY(u, -1.0, b));
112     PetscCall(VecNorm(u, NORM_INFINITY, &norm));
113     if (norm > tol) PetscCall(PetscPrintf(PETSC_COMM_WORLD, "%" PetscInt_FMT "-the %s numfact and solve: residual %g\n", k, factortype, (double)norm));
114   }
115 
116   /* Free data structures */
117   for (k = 0; k < num_numfac; k++) PetscCall(MatDestroy(&A[k]));
118   PetscCall(MatDestroy(&F));
119   PetscCall(ISDestroy(&perm));
120   PetscCall(ISDestroy(&iperm));
121   PetscCall(VecDestroy(&x));
122   PetscCall(VecDestroy(&b));
123   PetscCall(VecDestroy(&u));
124   PetscCall(PetscFinalize());
125   return 0;
126 }
127 
128 /*TEST
129 
130    test:
131 
132    test:
133       suffix: 2
134       args: -mat_solver_type superlu
135       requires: superlu
136 
137    test:
138       suffix: 3
139       nsize: 2
140       requires: mumps
141       args: -mat_solver_type mumps
142 
143    test:
144       suffix: 4
145       args: -mat_solver_type cusparse -mat_type aijcusparse -mat_factor_type {{lu cholesky ilu icc}separate output}
146       requires: cuda
147 
148 TEST*/
149