On Sun, 18 Dec 2005, Andries Hekstra wrote:

> Dear OCaml-list,
>
> I would like to thank several people of the ocaml list for constructive
> comments to the LDPC decoder program I posted earlier. I attached the
> revised Main.ml to this email.
>
> I noticed that when compiling Main.ml with -inline 100 -noassert -unsafe
> on a Pentium M, and I removing the unused argument messageCheckNodeOut of
> the function processCheckNode (resulting in Main2.ml), the computation
> time increases ca. 1.5-fold. I wonder whether this has to do with
> (un)boxing?

It's hard to say why OCaml somtimes produces the code the way it does,
it's even harder to predict how it will be intrepreted by different
"modern" CPUs. Here's a list of potentially useful links.

http://groups.google.com/group/fa.caml/browse_frm/thread/9c152a39b30c88d5/080df2a83aa2d878?lnk=st&q=group%3Afa.caml+author%3Amalc+faster&rnum=3#080df2a83aa2d878
http://groups.google.com/group/fa.caml/browse_thread/thread/624ed8213174b634/f5371b023927cc8e?lnk=st&q=group%3Afa.caml+author%3Amalc&rnum=15#f5371b023927cc8e
http://groups.google.com/group/fa.caml/browse_frm/thread/c9e40e633120a79/5f14ce70f4a35373?tvc=1&q=group%3Afa.caml+author%3Amalc#5f14ce70f4a35373
http://groups.google.com/group/comp.lang.functional/browse_frm/thread/4a6159c5beef192a/689140a6436f558f?lnk=st&q=malc+stand+corrected+ocaml&rnum=1#689140a6436f558f

-- 
mailto:malc@pulsesoft.com