#include <stdio.h>
#include <stdlib.h>
#include <sys/time.h>
#include <math.h>
#include <mpi.h>
#include <string.h>
char* help = "\n\
** Vector_Add_MPI.c ** \n\
Algoritm MPI de adunare a doua matrice \n\
\n\
Se compileaza: \n\
mpicc Vector_Add_MPI.c -o Vector_Add_MPI \n\
\n\
Se lanseaza cu parametrii n, p max_rep (n divizibil cu p) \n\
Implicit n = 1024, p = 1, max_rep = 1 \n\
mpirun -np p Matrix_Mult_MPI n max_rep \n\
\n\
Executie in mai multe noduri specificate in fisierul hosts:\n\
mpirun -hostfile hosts -np p Vector_Add_MPI n \n\
\n\
Fisierul hosts contine numele nodurilor (hostname) si numarul de procesoare (slots): \n\
hpc slots=1 \n\
compute-0-1 slots=1 \n\
compute-0-2 slots=1 \n\
compute-0-3 slots=1 \n";
int main(int argc, char* argv[])
{
int n = 1024; // Dimensiune matrice
int p = 1; // Nr procese MPI (p)
int max_rep = 1; // numar de repetari
int root = 0;
int rank; // Rang proces
int i, s, rep;
double psum, sum;
float *a,*b;
struct timeval t1, t2;
float tp;
char *fn = "Res_Reduction_MPI.txt";
FILE *fp;
// Citire parametri
if (argc >=2 && strcmp(argv[1],"-help") == 0) {
printf("%s\n",help);
return 0;
}
if (argc >= 2) n = atoi(argv[1]);
if (argc >= 3) max_rep = atoi(argv[2]);
// Testarea parametrilor
if (n < 2){
printf("Eroare: n < 2 \n");
exit(0);
}
if (max_rep < 1){
printf("Eroare: max_rep < 1 \n");
exit(0);
}
// Initializare MPI
MPI_Init (&argc,&argv);
MPI_Comm_rank (MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &p);
// Alocarea vectorilor
a = (float*)malloc(n*sizeof(float));
b = (float*)malloc(n*sizeof(float));
// Test n divizibil cu p
if ( n % p != 0) {
if (rank == root) printf("Eroare: n = %d nu este divizibil cu p = % d\n",n, p);
MPI_Finalize();
exit(0);
}
s = n / p;
// Initializare date
if (rank == root) {
for (i = 0; i < n; i++)
*(a+i) = 1;
*(b+i) = 1;
gettimeofday(&t1, NULL);
}
for (rep = 0; rep < max_rep; rep++) {
sum = 0; // Initializare sum
if (p == 1) {
// Executie secventiala
for (i = 0; i < n; i++)
sum += a[i]*b[i];
}
else {
// Executie paralela
if (rank == root && root == 0){
MPI_Scatter(a, s, MPI_FLOAT, MPI_IN_PLACE, s, MPI_FLOAT, root, MPI_COMM_WORLD);
MPI_Scatter(b, s, MPI_FLOAT, MPI_IN_PLACE, s, MPI_FLOAT, root, MPI_COMM_WORLD);
}
else{ MPI_Scatter(a, s, MPI_FLOAT, a, s, MPI_FLOAT, root, MPI_COMM_WORLD);
MPI_Scatter(b, s, MPI_FLOAT, b, s, MPI_FLOAT, root, MPI_COMM_WORLD);
}
psum = 0;
// Reducere locala in partitie
for (i = 0; i < s; i++) {
psum += a[i]*b[i];
}
// Reducere globala in root
MPI_Reduce(&psum, &sum, 1, MPI_DOUBLE, MPI_SUM,root, MPI_COMM_WORLD);
}
} // end rep
gettimeofday(&t2, NULL);
if (rank == root) {
tp = ((float)(t2.tv_sec - t1.tv_sec) + 0.000001*(t2.tv_usec - t1.tv_usec))/max_rep;
// Scriere in fisier
fp = fopen(fn, "a");
if (p == 1) fprintf(fp, "\n%f ",tp);
else fprintf(fp, "%f ",tp);
fclose(fp);
printf("%d rep, sum = %f n = %d, p = %d, tp = %f sec\n", rep, (float)sum, n, p, tp);
}
MPI_Finalize();
return 0;
}