What's the difficulty, exactly? You need some mundane float->bits and bits->float operations, which are commonly although not universally provided in strongly typed environments. The rest is integer ops.
That's not a terribly difficult proposition in a strongly-typed language -- in fact, it is even more performant than if you were to do it with tagged unions! The function "floatFromBits : float -> bits" has no execution overhead, and simply adds a type annotation to the semantic meaning of the identity. At execution time, there are no instructions emitted for it...
So, thanks for the well wishes, but in this case I don't think they're needed: any reasonable compiler should be able to do this without great difficulty.
It could just be a movd if the compiler backend people put the work in (and if they can't be arsed and implement it as a function call, that's nothing to do with the type system).
If the language supports union, then simply give your working variable the type of "union { float, long }". This directly expresses the idea that you're working with a piece of memory that is used for two purposes.
