#include<iostream>
#include<math.h>

int main()
{
 using namespace std;
 int i;
 float r,a[128*64];
 unsigned long long n,j,k;
 for(i=0;i<n;i++)
  cin>>a[i];
 j=n-1;
 for(k=j;k>=0;k--)
 {
  r=sqrt(a[k]);
  cout<<r<<"\n";
 }
 return 0;
}

You should try to increase the size of your array and perhaps use a double array over float.

> A size of the input stream does not exceed 256 KB

Minimum number of chars per number is 2 - "0 ". So you have to receive up to 128 K numbers. Your array is about 8 K numbers.

You shouldn't allocate such big arrays on stack. Use dynamic allocation or std::vector.

And another note.
> unsigned long long n,j,k;
> for(i=0;i<n;i++)
>  cin>>a[i];

How many times will this for iterate? Why?

Edited by author 15.12.2015 21:02

It's preferable to use a double precision floating point type over a single float. n is not given in the input data. I recommend to use std::vector to alloc it:

vector <double> input;
double aux;
while(cin>>aux) input.push_back(aux);