From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <weis@pauillac.inria.fr>
X-Original-To: caml-list@yquem.inria.fr
Delivered-To: caml-list@yquem.inria.fr
Received: from nez-perce.inria.fr (nez-perce.inria.fr [192.93.2.78])
	by yquem.inria.fr (Postfix) with ESMTP id B8EA8BB81
	for <caml-list@yquem.inria.fr>; Sat, 19 Nov 2005 17:02:42 +0100 (CET)
Received: from pauillac.inria.fr (pauillac.inria.fr [128.93.11.35])
	by nez-perce.inria.fr (8.13.0/8.13.0) with ESMTP id jAJG2gaa003947
	for <caml-list@yquem.inria.fr>; Sat, 19 Nov 2005 17:02:42 +0100
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 RAA10764 for <caml-list@pauillac.inria.fr>; Sat, 19 Nov 2005 17:02:41 +0100 (MET)
Received: from mx2.mail.ru (mx2.mail.ru [194.67.23.122])
	by concorde.inria.fr (8.13.0/8.13.0) with ESMTP id jAJG2f0s000745
	for <caml-list@inria.fr>; Sat, 19 Nov 2005 17:02:41 +0100
Received: from [80.237.161.183] (port=28433 helo=bely)
	by mx2.mail.ru with asmtp 
	id 1EdVAm-0006hn-00
	for caml-list@inria.fr; Sat, 19 Nov 2005 19:02:40 +0300
Received: from localhost (HELO BELY) [127.0.0.1]
	by bely (0.0.0.2) 
	with ESMTP (Classic Hamster Vr. 2.0 Build 2.0.0.1) ; Sat, 19 Nov 2005 18:49:49 +0300
To: caml-list@inria.fr
Subject: Bigarray access optimization?
From: Dmitry Bely <dbely@mail.ru>
Date: Sat, 19 Nov 2005 18:49:49 +0300
Message-ID: <4q68ljs2.fsf@mail.ru>
User-Agent: Gnus/5.1006 (Gnus v5.10.6) XEmacs/21.5 (chayote, windows-nt)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Posting-Agent: Hamster/2.0.0.1
X-Miltered: at nez-perce with ID 437F4CA2.000 by Joe's j-chkmail (http://j-chkmail.ensmp.fr)!
X-Miltered: at concorde with ID 437F4CA1.000 by Joe's j-chkmail (http://j-chkmail.ensmp.fr)!
X-Spam: no; 0.00; bigarray:01 bigarray:01 elt:01 ocamlopt:01 sizeof:01 compiler:01 indexing:01 reusing:01 bigarrays:01 -unsafe:01 sar:98 dmitry:01 dmitry:01 bely:02 bely:02 
X-Spam-Checker-Version: SpamAssassin 3.0.3 (2005-04-27) on yquem.inria.fr
X-Spam-Level: 
X-Spam-Status: No, score=0.0 required=5.0 tests=none autolearn=disabled 
	version=3.0.3

For the following code 

open Bigarray

let smooth (ba:(float, float32_elt, c_layout) Array2.t) =
    for i = 1 to (Array2.dim1 ba) - 2 do
      for j = 1 to (Array2.dim2 ba) - 2 do
        ba.{i,j} <- (ba.{i,j-1} +. ba.{i,j} +. ba.{i,j+1})/.3.0
      done
    done

ocamlopt generates the code that independently calculates offsets of
involved bigarray elements 4 times (each time using one "imul", one "add"
and two "sar" and several move instructions). Of course it's not necessary
- we just need to calculate offset once and then add or subtract
sizeof(float) to get ba.{i,j+1} and ba.{i,j-1} offsets accordingly. Can the
compiler perform such optimization? Say, testing if the difference between two
adjacent indexing expressions is a constant and then reusing the already
calculated offest if possible?

- Dmitry Bely

P.S. BTW, relaxing bounds checking for bigarrays when "-unsafe" option is
present seems to be quite easy. Is there a reason why it is not done?