/
inner.cpp
51 lines (40 loc) · 1.18 KB
/
inner.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
#include <stdlib.h>
#include <stdio.h>
#include <mpi.h>
double inner(long n, double* x, double* y, MPI_Comm comm) {
double local_prod = 0;
for (long i = 0; i < n; i++) local_prod += x[i] * y[i];
double prod = 0;
MPI_Reduce(&local_prod, &prod, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
return prod;
}
int main(int argc, char** argv) {
MPI_Init(&argc, &argv);
MPI_Comm comm = MPI_COMM_WORLD;
int mpirank, mpisize;
MPI_Comm_rank(comm, &mpirank);
MPI_Comm_size(comm, &mpisize);
long N = 100000000;
long N_local = N / mpisize;
long offset = N_local * mpirank;
double* x = (double*)malloc(N_local * sizeof(double));
double* y = (double*)malloc(N_local * sizeof(double));
for (long i = 0; i < N_local; i++) {
x[i] = 1.0 + offset + i;
y[i] = 2.0 / (1.0 + offset + i);
}
MPI_Barrier(comm);
double tt = MPI_Wtime();
double dot_prod = inner(N_local, x, y, comm);
tt = MPI_Wtime() - tt;
if (mpirank == 0) {
printf("inner-product = %f\n", dot_prod);
printf("time elapsed = %f\n", tt);
printf("%f GB/s\n", 2 * N * sizeof(double) / 1e9 / tt);
printf("%f Gflop/s\n", 2 * N / 1e9 / tt);
}
free(x);
free(y);
MPI_Finalize();
return 0;
}