New Approach Could Sink Floating Point Computation