/* Only the outmost loop can be parallelized. 
  
   The inner loop has loop carried anti data dependence.
   However, the loop is not parallelized so no race condition.

   Source: based on AutoPar's regression test.
 */
int n=100, m=100;
double b[100][100];

void foo()
{
  int i,j;
#pragma omp parallel for private(j)
  for (i=0;i<n;i++)
    for (j=1;j<m;j++) // Be careful about bounds of j
      b[i][j]=b[i][j-1];
}

int main()
{
  foo();
  return 0;
}