> I've been asked to research the use of Ocaml in high-performance
> computing applications on Mac OS X. I take this to mean (a) 64-bit
> applications (the client says so) and (b) compiled applications.

As you found out, ocamlopt currently cannot generate PPC64 code.
So, it's either bytecode-only in 64-bit mode, or native code in 32-bit
mode.

A PPC64 port of ocamlopt looks reasonably simple, but that will have
to wait until a Mac G5 lands here at INRIA.

> Make opt is not so happy. One example is that the assembler balks at
>
> _camlPervasives__55:
>     .long    _caml_int64_ops
>     .long    1018167296
>     .long    4372995238176751616
>     .data
>     .long    2303
>
> because the huge number should take the .quad directive, not .long.  But
> given that the huge number is 0x3cb0000000000000, I have bad  feelings
> that this is some badly-cleared or -shifted native integer.

No, no, it's a genuine 64-bit integer constant (see the definition of
epsilon_float in stdlib/pervasives.ml).

> In fact, my
> skimming of the compiler source suggests that the  compiler doesn't do
> native 64-bit pointers and integers.

The PowerPC back-end does not, but everything else in OCaml will
gladly handle 64-bit quantities on a 64-bit architecture (e.g. Alpha,
AMD64, IA64).

- Xavier Leroy