int main(int argc, char* argv[])
{
double *A, *B, *b, *y;
int n;
int my_rank, p;
int i;
/* Obtain number of rows and columns. We do not check for eroneous
input. */
n = atoi(argv[1]);
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &p);
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
/* Find how many rows per process. */
int *rows;
rows = (int*)Malloc(p, sizeof(int), MPI_COMM_WORLD, my_rank);
calcNumsPerProcess(n, p, rows);
/* Allocate memory. */
b = Malloc(n, sizeof(double), MPI_COMM_WORLD, my_rank);
if (my_rank == 0)
{
A = (double*)Malloc(n*n, sizeof(double), MPI_COMM_WORLD, my_rank);
y = (double*)Malloc(n, sizeof(double), MPI_COMM_WORLD, my_rank);
}
B = (double*)Malloc(rows[my_rank]*n, sizeof(double), MPI_COMM_WORLD, my_rank);
/* Generate matrix and vector */
if (my_rank == 0)
{
genMatrix(n, n, A);
genVector(n, b);
}
/* Distribute A */
int *displs;
int *sendcounts;
if (my_rank == 0)
{
displs = malloc(sizeof(int)*p);
sendcounts = malloc(sizeof(int)*p);
for (i=0; i<p; i++)
sendcounts[i] = rows[i]*n;
displs[0] = 0;
for (i=1; i<p; i++)
displs[i] = displs[i-1] + sendcounts[i-1];
}
MPI_Scatterv(A, sendcounts, displs, MPI_DOUBLE,
B, rows[my_rank]*n, MPI_DOUBLE, 0, MPI_COMM_WORLD);
/* Distribute b */
MPI_Bcast(b, n, MPI_DOUBLE, 0, MPI_COMM_WORLD);
double time = MPI_Wtime();
parallelMatrixTimesVector(rows[my_rank], n, B, b, y, 0, my_rank, p, MPI_COMM_WORLD);
time = MPI_Wtime()-time;
/* Collect the max time from all processes. */
double timerecv;
MPI_Reduce(&time,&timerecv, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
if (my_rank==0) {
printf("%d %d % .2e\n", p, n, timerecv);
}
if (my_rank==0){
free(sendcounts);
free(displs);
free(y);
}
free(A);
free(b);
free(rows);
MPI_Finalize();
return 0;
}
int main(int argc, char* argv[])
{
double *A, *B, *b, *y;
int num_rows, num_cols;
int my_rank, p;
int i;
/* Obtain number of rows and columns. We do not check for eroneous
input. */
num_rows = atoi(argv[1]);
num_cols= atoi(argv[2]);
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &p);
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
/* number of rows on my_rank */
int local_num_rows = NUM_ROWS(my_rank, p, num_rows);
/* Allocate memory */
b = Malloc(num_cols, sizeof(double), MPI_COMM_WORLD, my_rank);
if (my_rank == 0)
{
A = (double*)Malloc(num_rows*num_cols, sizeof(double), MPI_COMM_WORLD, my_rank);
y = (double*)Malloc(num_rows, sizeof(double), MPI_COMM_WORLD, my_rank);
}
B = (double*)Malloc(local_num_rows*num_cols, sizeof(double), MPI_COMM_WORLD, my_rank);
/* Generate matrix and vector */
if (my_rank == 0)
{
genMatrix(num_rows, num_cols, A);
genVector(num_cols, b);
}
/* Distribute A */
int *displs;
int *sendcounts;
if (my_rank == 0)
{
displs = malloc(sizeof(int)*p);
sendcounts = malloc(sizeof(int)*p);
sendcounts[0] = NUM_ROWS(0,p,num_rows)*num_cols;
displs[0] = 0;
for (i=1; i<p; i++)
{
displs[i] = displs[i-1] + sendcounts[i-1];
sendcounts[i] = NUM_ROWS(i,p,num_rows)*num_cols;
}
}
MPI_Scatterv(A, sendcounts, displs, MPI_DOUBLE,
B, local_num_rows*num_cols, MPI_DOUBLE, 0, MPI_COMM_WORLD);
/* Distribute b */
MPI_Bcast(b, num_cols, MPI_DOUBLE, 0, MPI_COMM_WORLD);
/* Multiply */
double time = MPI_Wtime();
parallelMatrixTimesVector(local_num_rows, num_cols, B, b, y, 0, my_rank, p, MPI_COMM_WORLD);
time = MPI_Wtime()-time;
/* Collect the max time from all processes. */
double timerecv;
MPI_Reduce(&time,&timerecv, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
if (my_rank==0)
printf("Computed with p = %d, m = %d, n = %d in % .2e seconds\n", p, num_rows, num_cols, timerecv);
if (my_rank == 0)
getResult(num_rows, num_cols, A, b, y);
if (my_rank==0){
free(sendcounts);
free(displs);
free(y);
free(A);
free(b);
}
free(B);
MPI_Finalize();
return 0;
}
