int main(int argc, char *argv[]) {
FILE *input_file, *output_file, *time_file;
if (argc < 4) {
fprintf(stderr, "Invalid input parameters\n");
fprintf(stderr, "Usage: %s inputfile outputfile timefile \n", argv[0]);
exit(1);
}
else {
input_file = fopen(argv[1], "r");
output_file = fopen(argv[2], "w");
time_file = fopen(argv[3], "w");
if (!input_file || !output_file || !time_file)
exit(1);
}
MM_typecode matcode;
if (mm_read_banner(input_file, &matcode) != 0) {
printf("Could not process Matrix Market banner.\n");
exit(1);
}
if (!mm_is_matrix(matcode) || !mm_is_real(matcode) || !mm_is_coordinate(matcode)) {
printf("Sorry, this application does not support ");
printf("Market Market type: [%s]\n", mm_typecode_to_str(matcode));
exit(1);
}
mtxMatrix inputMatrix, fullMatrix, LMatrix, UMatrix, UMatrixTranspose, MMatrix;
ReadMatrix(inputMatrix, input_file);
Timer timer;
getRowIndex(&inputMatrix, inputMatrix.RowIndex);
inputMatrix.RowIndex[inputMatrix.N] = inputMatrix.NZ;
int diagNum = 0;
for (int i = 0; i < inputMatrix.N; i++) {
for (int j = inputMatrix.RowIndex[i]; j < inputMatrix.RowIndex[i + 1]; j++) {
if (i == inputMatrix.Col[j]) diagNum++;
}
}
if (mm_is_symmetric(matcode)) {
InitializeMatrix(inputMatrix.N, 2 * inputMatrix.NZ - diagNum, fullMatrix);
TriangleToFull(&inputMatrix, &fullMatrix);
FreeMatrix(inputMatrix);
}
else {
fullMatrix = inputMatrix;
}
int *diag = new int[fullMatrix.N];
for (int i = 0; i < fullMatrix.N; i++) {
for (int j = fullMatrix.RowIndex[i]; j < fullMatrix.RowIndex[i + 1]; j++) {
if (i == fullMatrix.Col[j]) diag[i] = j;
}
}
// for (int i = 0; i < fullMatrix.N + 1; i++) {
// printf("RowIndex[%i] = %i\n", i, fullMatrix.RowIndex[i]);
// }
//
//
// for (int i = 0; i < fullMatrix.N; i++) {
// printf("input[%i]= %lf\n", i, fullMatrix.Value[inputMatrix.RowIndex[i]]);
// }
//
// for (int i = 0; i < fullMatrix.N; i++) {
// printf("diag[%i]= %d\n", i, diag[i]);
// }
timer.start();
ilu0(fullMatrix, fullMatrix.Value, diag);
LUmatrixSeparation(fullMatrix, diag, LMatrix, UMatrix);
Transpose(UMatrix, UMatrixTranspose);
Multiplicate(LMatrix, UMatrixTranspose, MMatrix);
timer.stop();
std::ofstream timeLog;
timeLog.open(argv[3]);
timeLog << timer.getElapsed();
WriteFullMatrix(MMatrix, output_file, matcode);
FreeMatrix(fullMatrix);
return 0;
}
int main(int argc, char** argv)
{
FILE *out;
coo_t *ilu = NULL, *A = NULL;
double *b = NULL;
ilu = (coo_t*) malloc (sizeof(coo_t));
A = (coo_t*) malloc (sizeof(coo_t));
int device = 0; //0 - CPU, 1 - GPU
int key_ilu = 0, key_A = 0, key_rhs = 0, dimb = 1, part = 1;
if (argc == 1)
{
fprintf(stderr,"Usage: %s -A [martix-market-filename] -ilu [martix-market-filename] -rhs [martix-market-filename]\n",argv[0]);
fprintf(stderr,"Optional: -d compute on GPU, -h compute on CPU(default), -b dim (where dim - number of phases on hydrodynamic model, default dim = 1)\n");
exit(1);
}
else
for (int i = 1; i < argc ; i++)
{
if (!strcmp(argv[i],"-ilu"))
{
key_ilu = 1;
if( read_coo_MM(argv[i+1], ilu) != 0)
exit(1);
}
if (!strcmp(argv[i],"-A"))
{
key_A = 1;
if( read_coo_MM(argv[i+1], A) != 0)
exit(1);
}
if (!strcmp(argv[i],"-rhs"))
{
key_rhs = 1;
if( read_coo_vector(argv[i+1], &b) != 0)
exit (1);
}
if (!strcmp(argv[i],"-d"))
device = 1;
if (!strcmp(argv[i],"-b"))
dimb = atoi(argv[i+1]);
if (!strcmp(argv[i],"-p"))
part = atoi(argv[i+1]);
}
if ( (key_A) && (key_rhs) )
{
if (key_ilu == 0)
{
ilu = allocCOO (A->nnz);
if( copyCOO (ilu,A) != 0) exit (1);
}
printf ("Load Matrix complete\n"); /*0-index based*/
}
else
{
printf ("Load Matrix Error\n");
exit(1);
}
csr_t* A_mat = allocCSR(A->nnz,A->n);
csr_t* ilu_mat = allocCSR(ilu->nnz,ilu->n);
coo2csr (A,A_mat);
//sortCSR(A_mat);
//printMatrix(A_mat);
coo2csr (ilu,ilu_mat);
//sortCSR(ilu_mat);
freeCOO(A);
freeCOO(ilu);
printf ("Convert 2csr complete\n");
//ilu0 and solve
double *x = (double*)calloc(A_mat->n, sizeof(double));
int key; //1 - block, 2 - usually
int nb;
if (device == 0)
{
if (part > 1)
{
key = 1;
nb = part;
}
else
key = 2;
}
else if (part == 1)
{
key = 3;
}
else
{
key = 4;
nb = part;
}
if ( key == 1 )
{
printf ("Usage block ilu0(%d)\n",nb);
/* Partition Matrix for ILU0 */
double time_partition = omp_get_wtime();
csr_t** bilu = partition(ilu_mat,nb,dimb);
printf ("time partition %lf\n",omp_get_wtime()-time_partition);
double time_bilu = omp_get_wtime();
int err = bilu0(bilu,nb);
printf ("time bilu0 %lf\n",omp_get_wtime()-time_bilu);
csr_zero2one_index(A_mat);
double time_solve = omp_get_wtime();
err = solve_bicgstab_block(A_mat,bilu,nb, b, x);
printf ("Time Solve %lf\n",omp_get_wtime()-time_solve);
for (int i = 0; i < nb; i++)
freeCSR(bilu[i]);
}
else if ( key == 2 )
{
printf ("Usage ilu0\n");
double time_ilu = omp_get_wtime();
int err = ilu0(ilu_mat);
printf ("Time ILU0 %lf\n",omp_get_wtime()-time_ilu);
csr_zero2one_index(A_mat);
double time_solve = omp_get_wtime();
err = solve_bicgstab(A_mat,ilu_mat, b, x);
printf ("Time Solve %lf\n",omp_get_wtime()-time_solve);
}
else if ( key == 3 )
{
printf ("Using block ilu0\n");
init();
int err = bicgstab_cusparse(A_mat,ilu_mat, b, x);
}
else if ( key == 4 )
{
printf ("Using block ilu0(%d)\n",nb);
/* Partition Matrix for ILU0 */
double time_partition = omp_get_wtime();
csr_t** bilu = partition(ilu_mat,nb,dimb);
csr_t** bA = partitionMatrix (A_mat,bilu,nb);
printf ("time partition %lf\n",omp_get_wtime()-time_partition);
//csr_zero2one_index(A_mat);
int err = bicgstab_fullblock_cusparse(bA, bilu, nb, b, x);
for (int i = 0; i < nb; i++)
{
freeCSR(bilu[i]);
freeCSR(bA[i]);
}
}
printVector(x,A_mat->n);
freeCSR(A_mat);
freeCSR(ilu_mat);
free(A);
free(ilu);
free(A_mat);
free(ilu_mat);
free (b);
free (x);
return 0;
}