From mboxrd@z Thu Jan  1 00:00:00 1970
Received: (from weis@localhost) by pauillac.inria.fr (8.7.6/8.7.3) id JAA21598 for caml-red; Fri, 21 Jul 2000 09:26:06 +0200 (MET DST)
Received: from concorde.inria.fr (concorde.inria.fr [192.93.2.39]) by pauillac.inria.fr (8.7.6/8.7.3) with ESMTP id WAA21141 for <caml-list@pauillac.inria.fr>; Wed, 19 Jul 2000 22:13:44 +0200 (MET DST)
Received: from pauillac.inria.fr (pauillac.inria.fr [128.93.11.35])
	by concorde.inria.fr (8.10.0/8.10.0) with ESMTP id e6JKDan24448;
	Wed, 19 Jul 2000 22:13:36 +0200 (MET DST)
Received: (from xleroy@localhost) by pauillac.inria.fr (8.7.6/8.7.3) id WAA26875; Wed, 19 Jul 2000 22:13:36 +0200 (MET DST)
Message-ID: <20000719221336.34904@pauillac.inria.fr>
Date: Wed, 19 Jul 2000 22:13:36 +0200
From: Xavier Leroy <Xavier.Leroy@inria.fr>
To: Caml Mailing List <caml-list@inria.fr>, lindig@eecs.harvard.edu
Subject: Re: IEEE floating point emulation library?
References: <20000711213139.A339@eecs.harvard.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.89.1
In-Reply-To: <20000711213139.A339@eecs.harvard.edu>; from Christian Lindig on Tue, Jul 11, 2000 at 09:31:39PM -0400
Sender: weis@pauillac.inria.fr

> For a compiler project I am looking for an IEEE floating point
> emulation.  I have already checked http://caml.inria.fr/hump.html and
> the OCaml link database for such a libray.  Does anybody has written
> something like this for OCaml?

This sounds hard.  IEEE floating-point arithmetic is quite intricate.
However:

> To answer the obvious question, why not use the built in float data
> type:  I need access to the representation at the bit level.  Some
> code that decodes a float value into a bit vector (int32/int64) and
> back would be also helpful.

You can do this easily using the following piece of C code:

  #include <caml/mlvalues.h>
  #include <caml/alloc.h>

  value unpack_float(value f)
  {
    union { double d; int64 i; } buffer;
    buffer.d = Double_val(f);
    return copy_int64(buffer.i);
  }

  value pack_float(value i)
  {
    union { double d; int64 i; } buffer;
    buffer.i = Int64_val(i);
    return copy_double(buffer.d);
  }

and the following external declarations:

  external unpack_float: float -> int64 = "unpack_float"
  external pack_float: int64 -> float = "pack_float"

>>From the int64 representation, you can easily access every bit of the
float representation.  For single-precision floats and int32, substitute
"double" by "float" and "int64" by "int32" in the code above.

Hope this helps.

- Xavier Leroy