Comment author: administrator

I had the problem that floats got corrupted using the bytecode
runtime on the ARM :
Later Xavier Leroy that it tested this on the skiffcluster (handhelds.org).

After some investigation I discovered that yes the skiffcluster5.handhelds.org
ocamlc compiler from Xavier worked on the foo.ml. But at the very same
time "float", the same program but compiled on my machine don't work as I put
it before.

Ah, you didn't say you were cross-compiling. This is a known problem
with ARM floating-point arithmetic: its double-precision floats are
neither little-endian nor big-endian. That is:

                    representation of 1.0

little-endian 00 00 00 00 00 00 f0 3f
big-endian 3f f0 00 00 00 00 00 00
ARM 00 00 f0 3f 00 00 00 00

The Caml bytecode interpreter can convert between little-endian and
big-endian formats (e.g. if you compiled on a l-e machine and run on a
b-e machine). But it cannot convert to/from the ARM format.

So, I'm afraid you have to compile on the ARM itself if your program
contains float literals.

• Xavier Leroy

Comment author: administrator

Hi,

Ah, you didn't say you were cross-compiling. This is a known problem
Désolé je pensais que ça n'avait aucune incidence vu que il me semblait
le bytecode était complètement indépendant du système compilé.
with ARM floating-point arithmetic: its double-precision floats are
neither little-endian nor big-endian. That is:

                   representation of 1.0

little-endian 00 00 00 00 00 00 f0 3f
big-endian 3f f0 00 00 00 00 00 00
ARM 00 00 f0 3f 00 00 00 00

The Caml bytecode interpreter can convert between little-endian and
big-endian formats (e.g. if you compiled on a l-e machine and run on a
b-e machine). But it cannot convert to/from the ARM format.
Serait-il difficile de l'implémenter ? Bon d'accord tout est possible,
mais ça reste relatif. Si vous n'avez pas l'intention de le faire
dans un futur immédiat, pourriez-vous me donner quelques pistes
pour l'implémenter ? Je serais plus que content d'essayer de contribuer
à Ocaml de cette manière.

So, I'm afraid you have to compile on the ARM itself if your program
contains float literals.
Ça réduit considérablement l'intérêt du cross-compiler et vu la vitesse
de l'iPAQ c'est bien plus pratique de cross-compiler.

Cordialement,

--
Christian Gillot cgillot@gruposbd.com
GNU/Linux developer

Comment author: administrator

Fixed 2002-02-19 by XL