The difficulty here lies in the precision of floating point numbers.

You can:
 - read each line as INTEGER DOT INTEGER then multiply this
   by 10, sum everything on a 64bits integers
   
 - use long double