Compilation of single-/double-float arithmetic to immediate byte code

[ehuelsmann]

Currently, the following code

"
(lambda (x)

(declare (double-float x))
(+ 3.0e0

(* 12.0e0 x)))

"

disassembles to a call to a LispObject function for + and one for *.

However, this code could be more efficient if only 1 new object would be required. This can be achieved by unboxing X, then doing the math in Java byte code and boxing the result before it's returned.

[ehuelsmann]

The example above has been realised, hence, closing this issue - further more, all intermediate results which don't get stored into registers won't be boxed now. However, local variables of float types don't get stored unboxed yet. Opening a new issue for it.