From mboxrd@z Thu Jan  1 00:00:00 1970
Received: (from majordomo@localhost) by pauillac.inria.fr (8.7.6/8.7.3) id XAA32536; Thu, 22 Jul 2004 23:05:25 +0200 (MET DST)
X-Authentication-Warning: pauillac.inria.fr: majordomo set sender to owner-caml-list@pauillac.inria.fr using -f
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 XAA00408 for <caml-list@pauillac.inria.fr>; Thu, 22 Jul 2004 23:05:24 +0200 (MET DST)
Received: from mail3.bluewin.ch (mail3.bluewin.ch [195.186.1.75])
	by concorde.inria.fr (8.12.10/8.12.10) with ESMTP id i6ML5NSH015355
	for <caml-list@inria.fr>; Thu, 22 Jul 2004 23:05:23 +0200
Received: from [10.0.1.2] (81.62.93.107) by mail3.bluewin.ch (Bluewin AG 7.0.030.2)
        id 40ED9D270019FD59 for caml-list@inria.fr; Thu, 22 Jul 2004 21:05:22 +0000
Mime-Version: 1.0 (Apple Message framework v618)
Content-Transfer-Encoding: 7bit
Message-Id: <146D181C-DC23-11D8-8ADE-000393DBC266@epfl.ch>
Content-Type: text/plain; charset=US-ASCII; format=flowed
To: caml-list@inria.fr
From: =?ISO-8859-1?Q?Daniel_B=FCnzli?= <daniel.buenzli@epfl.ch>
Subject: [Caml-list] Interesting optimization
Date: Thu, 22 Jul 2004 23:07:03 +0200
X-Mailer: Apple Mail (2.618)
X-Miltered: at concorde with ID 41002C13.000 by Joe's j-chkmail (http://j-chkmail.ensmp.fr)!
X-Loop: caml-list@inria.fr
X-Spam: no; 0.00; obsessed:01 checksum:01 rewrote:01 ocamlopt:01 0.010:01 int's:01 int:01 int:01 rewritten:02 sys:03 sys:03 usefull:03 let:04 let:04 parse:04 
Sender: owner-caml-list@pauillac.inria.fr
Precedence: bulk

Hello,

I usually try not to be too much obsessed with speed, but I had the 
following interesting experience. While rearanging some checksum code I 
thought that I had rewritten it in a more efficient way. However I 
turned out not to be the case.

I can boil the example to the following (n.b. loops don't compute 
anything usefull). Basically I rewrote update into update'.

--- test.ml ---
let update c =
   let c' = ref !c in
   for n = 0 to max_int do
     c' := !c' land 0xff
   done;
   c := !c'

let update' c =
   for n = 0 to max_int do
     c := !c land 0xff;
   done

let compute use_ref =
   let x = ref 2 in
   if use_ref then update' x else update x;
   print_int !x

let main () =
   let use_ref = ref false in
   let args = [("-ref", (Arg.Set use_ref), "use reference directly")] in
   Arg.parse args (fun _ -> ()) "";
   compute !use_ref

let () = main ()
---------------

 > ocamlopt -o test.opt test.ml
 > time ./test.opt
2
real    0m3.500s
user    0m3.230s
sys     0m0.010s
 > time ./test.opt -ref
2
real    0m7.599s
user    0m7.550s
sys     0m0.030s

The few that I can read of ppc assembly tells me that in update the 
value of c' is directly stored in a register whearas update' accesses 
memory on each iteration.

Note that this is not restricted to int's, it occured to me with an 
int32. I guess it should work with anything that gets into a register.

Cheers,

Daniel

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners