//#define Debug
int main(int argc, char **argv)
{
if(argc != 3)
{
printf("Usage: test {N}\n");
exit(-1);
}
size_t N = atoi(argv[1]);
size_t npro = atoi(argv[2]);
omp_set_num_threads(npro);
int iam = 0, np = 1;
#pragma omp parallel private(iam, np)
{
np = omp_get_num_threads();
iam = omp_get_thread_num();
#ifdef Debug
printf("Hello from thread %d out of %d\n", iam, np);
#endif
}
struct timeval tv1,tv2;
// generate a random matrix.
printf("Size is %d,numP is %d\n",N,npro);
int *mat = (int*)malloc(sizeof(int)*N*N);
GenMatrix(mat, N);
// compute the reference result.
int *ref = (int*)malloc(sizeof(int)*N*N);
memcpy(ref, mat, sizeof(int)*N*N);
gettimeofday(&tv1,NULL);
ST_APSP(ref, N);
gettimeofday(&tv2,NULL);
printf("Sequential time = %ld usecs\n",
(tv2.tv_sec-tv1.tv_sec)*1000000+tv2.tv_usec-tv1.tv_usec);
// compute your results
int *result = (int*)malloc(sizeof(int)*N*N);
memcpy(result, mat, sizeof(int)*N*N);
// ST_APSP(result, N);
gettimeofday(&tv1,NULL);
OMP_APSP(result,N);
gettimeofday(&tv2,NULL);
printf("OpenMp time = %ld usecs\n",
(tv2.tv_sec-tv1.tv_sec)*1000000+tv2.tv_usec-tv1.tv_usec);
// compare your result with reference result
if(CmpArray(result, ref, N*N))
printf("Your result is correct.\n");
else
printf("Your result is wrong.\n");
}
int main(int argc, char **argv)
{
if(argc != 2)
{
printf("Usage: test {N}\n");
exit(-1);
}
size_t N = atoi(argv[1]);
// matrix related variables
int *mat, *ref, *result, *part, *k_row;
int rows, k, root;
int i, j, vij, vik, vkj;
int npes, rank;
struct timeval tv1, tv2;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &npes);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if(rank == 0){
// generate matrix and compute sequential
mat = (int*)malloc(sizeof(int)*N*N);
ref = (int*)malloc(sizeof(int)*N*N);
result = (int*)malloc(sizeof(int)*N*N);
GenMatrix(mat, N);
memcpy(ref, mat, sizeof(int)*N*N);
gettimeofday(&tv1, NULL);
ST_APSP(ref, N);
gettimeofday(&tv2, NULL);
printf("Sequential: %ld usecs\n",(tv2.tv_sec-tv1.tv_sec)*1000000+tv2.tv_usec-tv1.tv_usec);
gettimeofday(&tv1, NULL);
}
// scatter the matrix
rows = N/npes;
part = (int*)malloc(sizeof(int)*N*rows);
MPI_Scatter(mat, N*rows, MPI_INT, part, N*rows, MPI_INT, 0, MPI_COMM_WORLD);
// parallel computing
k_row = (int*)malloc(sizeof(int)*N);
for(k = 0; k < N; k++){
root = k/rows;
if(rank == root){
for(i = 0; i < N; i ++){
*(k_row+i) = *(part + N*(k-rows*root) + i);
}
}
MPI_Bcast(k_row, N, MPI_INT, root, MPI_COMM_WORLD);
for(i = 0; i < rows; i++){
for(j = 0; j < N; j++){
vij = *(part + N*i + j);
vik = *(part + N*i + k);
vkj = *(k_row + j);
if(vik != -1 && vkj != -1){
if(vij == -1 || vij > vik+vkj)
*(part + N*i + j) = vik + vkj;
}
}
}
}
// gather the matrix
MPI_Gather(part, N*rows, MPI_INT, result, N*rows, MPI_INT, 0, MPI_COMM_WORLD);
//compare your result with reference result
if(rank == 0){
gettimeofday(&tv2, NULL);
printf("Parallel: %ld usecs\n",(tv2.tv_sec-tv1.tv_sec)*1000000+tv2.tv_usec-tv1.tv_usec);
// printmat(mat, N);
// printmat(ref, N);
// printmat(result, N);
if(CmpArray(result, ref, N*N))
printf("Your result is correct.\n");
else
printf("Your result is wrong.\n");
}
// free memory
if(rank == 0){
free(mat);
free(ref);
free(result);
}
free(part);
free(k_row);
MPI_Finalize();
}
int main(int argc, char **argv) {
MPI_Init(NULL, NULL);
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
MPI_Comm_size(MPI_COMM_WORLD, &total_rank);
if(argc != 2) {
printf("Usage: test {mat_size}\n");
exit(-1);
}
size_t mat_size = atoi(argv[1]);
if (mat_size % total_rank) {
for (int i = 0; i < 20; i++)
printf("~");
printf("\n");
printf("Using MPT_*, only support divisible number of vertex...\n");
for (int i = 0; i < 20; i++)
printf("~");
printf("\n");
exit(-1);
}
int* mat;
int* ref;
int* result;
if (my_rank == 0) {
for (int i = 0; i < 20; i++)
printf("~");
printf("\n");
printf("Using MPT_*, only support divisible number of vertex...\n");
printf("Input size: %ld\n", mat_size);
printf("Total process: %d\n", total_rank);
prepare_data(&mat, &ref, &result, mat_size);
// start the timer
gettimeofday(&timer_parallel, NULL);
}
MPI_Barrier(MPI_COMM_WORLD);
// know the set of rows I am working on according to my_rank
int rows_in_charge = mat_size / total_rank;
// should this be included in the timer?
int* my_rows = (int*)malloc(sizeof(int) * mat_size * rows_in_charge); //rows the current process have
int* k_to_j = (int*)malloc(sizeof(int) * mat_size); // the vertical (column)
if (my_rank == 0)
gettimeofday(&timer_comm, NULL);
// divide the matrix for each process
// send rows to each process using scatter, sendbuf:*result, recvbuf:*my_rows
int sendrecvcount = mat_size * rows_in_charge;
MPI_Scatter(
result,
sendrecvcount,
MPI_INT,
my_rows,
sendrecvcount,
MPI_INT,
0,
MPI_COMM_WORLD);
if (my_rank == 0)
time_comm += get_time_and_replace(&timer_comm);
// preprocess_graph(my_rows, rows_in_charge, mat_size);
for (int k = 0; k < mat_size; k++) {
if (my_rank == 0)
gettimeofday(&timer_comm, NULL);
// broadcast k-th row to other process if I am the owner
int owner_of_k_row = k / rows_in_charge;
if (my_rank == owner_of_k_row)
memcpy(k_to_j, my_rows + mat_size * (k % rows_in_charge), sizeof(int) * mat_size);
MPI_Bcast(k_to_j, mat_size, MPI_INT, owner_of_k_row, MPI_COMM_WORLD);
if (my_rank == 0)
time_comm += get_time_and_replace(&timer_comm);
for (int i = 0; i < rows_in_charge; i++) {
for (int j = 0; j < mat_size; j++) {
int ij = i * mat_size + j;
int ik = i * mat_size + k;
// if (my_rows[ij] > ikj)
// my_rows[ij] = ikj;
if (my_rows[ik] != -1 && k_to_j[j] != -1) {
int ikj = my_rows[ik] + k_to_j[j];
if (my_rows[ij] == -1 || my_rows[ij] > ikj)
my_rows[ij] = ikj;
}
}
}
}
if (my_rank == 0)
gettimeofday(&timer_comm, NULL);
// collect result to process 0
MPI_Gather(
my_rows,
sendrecvcount,
MPI_INT,
result,
sendrecvcount,
MPI_INT,
0,
MPI_COMM_WORLD);
if (my_rank == 0)
time_comm += get_time_and_replace(&timer_comm);
if (my_rank == 0) {
//stop the timer
time_used_parallel = get_time_and_replace(&timer_parallel);
printf("Time used (parallel ): %8ld usecs\n", time_used_parallel);
printf("Time used (parallel ) comm : %6ld usecs (%2.3lf%%) \n", time_comm, time_comm / (double)time_used_parallel * 100);
printf("Speed up (sequential / parallel): %.3lf\n", time_used_sequential / (double)time_used_parallel);
//compare your result with reference result
if(CmpArray(result, ref, mat_size * mat_size))
printf("Your result is correct.\n");
else
printf("Your result is wrong.\n");
for (int i = 0; i < 20; i++)
printf("~");
printf("\n");
}
// Finalize the MPI environment.
MPI_Finalize();
}
int main(int argc, char *argv[]) {
if(argc != 2)
{
printf("Missing Argument\n");
exit(-1);
}
int i, j;
int *mat, *ref;
int P, myrank;
size_t N = atoi(argv[1]); //matrix size
struct timeval tv1,tv2;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &P);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
if(myrank == 0) {
// generate a random matrix.
printf("input Size is %d; number of process is %d \n",N,P);
mat = (int*)malloc(sizeof(int)*N*N);
GenMatrix(mat, N);
// compute the reference result.
ref = (int*)malloc(sizeof(int)*N*N);
memcpy(ref, mat, sizeof(int)*N*N);
gettimeofday(&tv1, NULL);
ST_APSP(ref, N);
gettimeofday(&tv2, NULL);
printf("Sequential time = %ld usecs\n",
(tv2.tv_sec-tv1.tv_sec)*1000000+tv2.tv_usec-tv1.tv_usec);
// compute your results
}
int strip = N/P;
int *part = (int*)malloc(sizeof(int)*N*strip); /*array holding the part for this processor*/
// Scatter data to all processors
MPI_Barrier(MPI_COMM_WORLD);
if (myrank == 0) {
gettimeofday(&tv1,NULL);
}
MPI_Scatter(mat, N*strip, MPI_INT, part, N*strip, MPI_INT, 0, MPI_COMM_WORLD);
// Compute matrix in parallel
MT_APSP (part, MPI_COMM_WORLD, myrank, N, P);
//Gather the results
MPI_Gather(part, N*strip, MPI_INT, mat, N*strip, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Barrier(MPI_COMM_WORLD);
if(myrank == 0) {
gettimeofday(&tv2, NULL);
printf("Parallel time = %ld usecs\n\n",
(tv2.tv_sec-tv1.tv_sec)*1000000+tv2.tv_usec-tv1.tv_usec);
#ifdef test
// compare your result with reference result
if(CmpArray(mat, ref, N*N))
printf("Your result is correct.\n");
else
printf("Your result is wrong.\n");
#endif
free(mat);
free(ref);
}
free(part);
MPI_Finalize();
return 0;
}