MPI_Wtime() returns strange numbers - mpi

Question

When I use MPI_Wtime() to record the computation time, it returns several weird numbers, which seems not correct. My code is very simple:
#include <stdio.h>
#include "mpi.h"
int main(int argc,char *argv[])
{
int i,sum,total;
int size,rank,start,end,sumGlobal;
double t1,t2;
total = 10000;
t1 = MPI_Wtime();
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&size);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
start = rank*(total/size)+1;
if (rank==(size-1))
{
end = total;
}
else
{
end = start+(total/size)-1;
}
sumGlobal = 0;
sum = 0;
for (i=start;i<=end;i++)
{
sum += i;
}
MPI_Reduce(&sum,&sumGlobal,1,MPI_INT,MPI_SUM,0,MPI_COMM_WORLD);
t2 = MPI_Wtime();
printf("rank = %d,sum = %d,sumGlobal = %d,time=%f\n",rank,sum,sumGlobal,t2-t1);
MPI_Finalize();
return 0;
}
The running result is (use 16 cores):
rank = 0,sum = 195625,sumGlobal = 50005000,time = 1520044462.143606
rank = 1,sum = 586250,sumGlobal = 0,time = 1520044462.143662
rank = 3,sum = 1367500,sumGlobal = 0,time = 1520044462.143665
rank = 4,sum = 1758125,sumGlobal = 0,time = 1520044462.143610
rank = 5,sum = 2148750,sumGlobal = 0,time = 1520044462.143651
rank = 6,sum = 2539375,sumGlobal = 0,time = 1520044462.143608
rank = 7,sum = 2930000,sumGlobal = 0,time = 1520044462.143604
rank = 9,sum = 3711250,sumGlobal = 0,time = 1520044462.143647
rank = 10,sum = 4101875,sumGlobal = 0,time = 1520044462.143655
rank = 11,sum = 4492500,sumGlobal = 0,time = 1520044462.143606
rank = 12,sum = 4883125,sumGlobal = 0,time = 1520044462.143614
rank = 13,sum = 5273750,sumGlobal = 0,time = 1520044462.143672
rank = 14,sum = 5664375,sumGlobal = 0,time = 1520044462.143643
rank = 15,sum = 6055000,sumGlobal = 0,time = 1520044462.143605
rank = 2,sum = 976875,sumGlobal = 0,time = 1520044462.143665
rank = 8,sum = 3320625,sumGlobal = 0,time = 1520044462.143652
As you can see, the "time" values are very strange. So how to fix it? Thank you.

Answer 1

Your program is incorrect with respect to the MPI standard :
you simply cannot use MPI_Wtime() before MPI_Init() (nor after MPI_Finalize() fwiw).
Moving the t1 initialization should fix the issue.