C parallel implementation of Gauss elimination with MPI [closed] - c

Question

Closed. This question needs debugging details. It is not currently accepting answers.
Edit the question to include desired behavior, a specific problem or error, and the shortest code necessary to reproduce the problem. This will help others answer the question.
Closed 7 years ago.
Improve this question
I'm very new to MPI and I was asked to write a C parallel implementation for Gauss elimination (without pivoting).
I gave it a try (I used a row-wise decomposition) but my code doesn't work. I am hoping someone can give me some pointers here. I've been looking for what's wrong for few days already without success :(
Thank you in advance !
#include<stdio.h>
#include <stdlib.h>
#include <time.h>
#include <mpi.h>
int main(int argc, char **argv)
{
MPI_Init(&argc, &argv);
int i,j,k;
int map[500];
float A[500][500],b[500],c[500],x[500],sum=0.0;
double range=1.0;
int n=3;
int rank, nprocs;
clock_t begin1, end1, begin2, end2;
MPI_Status status;
MPI_Comm_rank(MPI_COMM_WORLD, &rank); /* get current process id */
MPI_Comm_size(MPI_COMM_WORLD, &nprocs); /* get number of processes */
//////////////////////////////////////////////////////////////////////////////////
if (rank==0)
{
for (i=0; i<n; i++)
{
for (j=0; j<n; j++)
A[i][j]=range*(1.0-2.0*(double)rand()/RAND_MAX);
b[i]=range*(1.0-2.0*(double)rand()/RAND_MAX);
}
printf("\n Matrix A (generated randomly):\n");
for (i=0; i<n; i++)
{
for (j=0; j<n; j++)
printf("%9.6lf ",A[i][j]);
printf("\n");
}
printf("\n Vector b (generated randomly):\n");
for (i=0; i<n; i++)
printf("%9.6lf ",b[i]);
printf("\n\n");
}
//////////////////////////////////////////////////////////////////////////////////
begin1 =clock();
MPI_Bcast (A,n*n,MPI_DOUBLE,0,MPI_COMM_WORLD);
MPI_Bcast (b,n,MPI_DOUBLE,0,MPI_COMM_WORLD);
for(i=0; i<n; i++)
{
map[i]= i % nprocs;
}
for(k=0;k<n;k++)
{
MPI_Bcast (&A[k][k],n-k,MPI_DOUBLE,map[k],MPI_COMM_WORLD);
MPI_Bcast (&b[k],1,MPI_DOUBLE,map[k],MPI_COMM_WORLD);
for(i= k+1; i<n; i++)
{
if(map[i] == rank)
{
c[i]=A[i][k]/A[k][k];
}
}
for(i= k+1; i<n; i++)
{
if(map[i] == rank)
{
for(j=0;j<n;j++)
{
A[i][j]=A[i][j]-( c[i]*A[k][j] );
}
b[i]=b[i]-( c[i]*b[k] );
}
}
}
end1 = clock();
//////////////////////////////////////////////////////////////////////////////////
begin2 =clock();
if (rank==0)
{
x[n-1]=b[n-1]/A[n-1][n-1];
for(i=n-2;i>=0;i--)
{
sum=0;
for(j=i+1;j<n;j++)
{
sum=sum+A[i][j]*x[j];
}
x[i]=(b[i]-sum)/A[i][i];
}
end2 = clock();
}
//////////////////////////////////////////////////////////////////////////////////
if (rank==0)
{
printf("\nThe solution is:");
for(i=0;i<n;i++)
{
printf("\nx%d=%f\t",i,x[i]);
}
printf("\n\nLU decomposition time: %f", (double)(end1 - begin1) / CLOCKS_PER_SEC);
printf("\nBack substitution time: %f\n", (double)(end2 - begin2) / CLOCKS_PER_SEC);
}
return(0);
MPI_Finalize();
}
And this is the error I'm getting:
mpirun has exited due to process rank 1 with PID XXXX on node XXXX exiting without calling "finalize". This may have caused other processes in the application to be terminated by signals sent by mpirun (as reported here).

Answer 1

As noticed by High Performance Mark, add MPI_Finalize() before return(0). This code will run without prompting any problem...But the result will still be uncorrect. In parallel, it will print nan as being the result, which is false.
The problem comes from MPI_Bcast(A,n*n,MPI_DOUBLE,...). A is defined as float A[500][500].
You need to broadcast the pointer to the first element &A[0][0], not the pointer to the pointer to the first element.
If you send n*n elements (n=3), you will send A[0][0],...,A[0][8] and A[1][1] will be left uninitialized. This could cause wrong results, such as nan. For the shake of simplicity (laziness...), you may change for 500*500.
MPI_DOUBLE corresponds to double precision... Solution is either to change for double A[500][500] or MPI_Bcast(&A[0][0],500*500,MPI_FLOAT,...). Do the same thing for b.
This deterministic use of rand() is really useful for debugging purpose...Do not forget to use srand() to seed your random generator !
EDIT : here is the code :
#include<stdio.h>
#include <stdlib.h>
#include <time.h>
#include <mpi.h>
int main(int argc, char **argv)
{
MPI_Init(&argc, &argv);
int i,j,k;
int map[500];
double A[500][500],b[500],c[500],x[500],sum=0.0;
double range=1.0;
int n=3;
int rank, nprocs;
clock_t begin1, end1, begin2, end2;
MPI_Status status;
MPI_Comm_rank(MPI_COMM_WORLD, &rank); /* get current process id */
MPI_Comm_size(MPI_COMM_WORLD, &nprocs); /* get number of processes */
//////////////////////////////////////////////////////////////////////////////////
if (rank==0)
{
for (i=0; i<n; i++)
{
for (j=0; j<n; j++)
A[i][j]=range*(1.0-2.0*(double)rand()/RAND_MAX);
b[i]=range*(1.0-2.0*(double)rand()/RAND_MAX);
}
printf("\n Matrix A (generated randomly):\n");
for (i=0; i<n; i++)
{
for (j=0; j<n; j++)
printf("%9.6lf ",A[i][j]);
printf("\n");
}
printf("\n Vector b (generated randomly):\n");
for (i=0; i<n; i++)
printf("%9.6lf ",b[i]);
printf("\n\n");
}
//////////////////////////////////////////////////////////////////////////////////
begin1 =clock();
MPI_Bcast (&A[0][0],500*500,MPI_DOUBLE,0,MPI_COMM_WORLD);
MPI_Bcast (b,n,MPI_DOUBLE,0,MPI_COMM_WORLD);
for(i=0; i<n; i++)
{
map[i]= i % nprocs;
}
for(k=0;k<n;k++)
{
MPI_Bcast (&A[k][k],n-k,MPI_DOUBLE,map[k],MPI_COMM_WORLD);
MPI_Bcast (&b[k],1,MPI_DOUBLE,map[k],MPI_COMM_WORLD);
for(i= k+1; i<n; i++)
{
if(map[i] == rank)
{
c[i]=A[i][k]/A[k][k];
}
}
for(i= k+1; i<n; i++)
{
if(map[i] == rank)
{
for(j=0;j<n;j++)
{
A[i][j]=A[i][j]-( c[i]*A[k][j] );
}
b[i]=b[i]-( c[i]*b[k] );
}
}
}
end1 = clock();
//////////////////////////////////////////////////////////////////////////////////
begin2 =clock();
if (rank==0)
{
x[n-1]=b[n-1]/A[n-1][n-1];
for(i=n-2;i>=0;i--)
{
sum=0;
for(j=i+1;j<n;j++)
{
sum=sum+A[i][j]*x[j];
}
x[i]=(b[i]-sum)/A[i][i];
}
end2 = clock();
}
//////////////////////////////////////////////////////////////////////////////////
if (rank==0)
{
printf("\nThe solution is:");
for(i=0;i<n;i++)
{
printf("\nx%d=%f\t",i,x[i]);
}
printf("\n\nLU decomposition time: %f", (double)(end1 - begin1) / CLOCKS_PER_SEC);
printf("\nBack substitution time: %f\n", (double)(end2 - begin2) / CLOCKS_PER_SEC);
}
MPI_Finalize();
return(0);
}

Answer 2

I'm not much of a C programmer but it looks to me as if you have probably called return prematurely. Specifically you have called it before MPI_Finalize(). Try swapping the order of the statements. Or even dropping the return altogether.