OCaml floats boxed or un-boxed?

Question

I know that floating point numbers are normally boxed in OCaml, but my difficulty is that word: normally. When are they not boxed? And, if they aren't boxed, how are they represented so the runtime recognizes them as different from ints or pointers?

I've found http://caml.inria.fr/pub/old_caml_site/ocaml/numerical.html, which lists certain times when floats aren't boxed, but it's 11 years old so I don't know if it's still up-to-date, and it doesn't explain HOW they're represented when not boxed.

I'm new to OCaml, so sorry if this is a dumb noob question. Thanks!

Answer 1

Floats are unboxed when in an array and when in a record all of whose fields are float. There is a special tag for these cases that marks the collection as containing unboxed floats.

This is described in Section 19.3 of the OCaml manual.

Answer 2

Floats are unboxed in two kinds of situation:

• When there are only floats in an array (even in polymorphic functions), or in a record (this time, only when all fields are determined as floats at compile time)

• During sequences of float operations: the compiler will notice that it is not useful to box a float just before unboxing it for the next operation using it, so the float remains unboxed. In:

  let x = 
    let y = a +. b in
    y *. c
  

  The value y is not going to be boxed. This optimization is also done on int32, int64 and nativeint, so that they have actually very good performances in heavy computations.