Linpack Benchmark for Java

We forced all floating-point operations to comply with the strict semantics.

We expected that the fscale method shows better results,
         because the daxpy() loop contains only an add. and an multiplication.
Conversely, the multiplication (fmul) method was better...