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 PAA07463; Mon, 19 Aug 2002 15:13:27 +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 PAA07396 for <caml-list@pauillac.inria.fr>; Mon, 19 Aug 2002 15:13:26 +0200 (MET DST)
Received: from comtv.ru (comtv.ru [217.10.32.4])
	by concorde.inria.fr (8.11.1/8.11.1) with ESMTP id g7JDDPT10764
	for <caml-list@inria.fr>; Mon, 19 Aug 2002 15:13:26 +0200 (MET DST)
Received: from [10.0.66.9] ([10.0.66.9] verified)
  by comtv.ru (CommuniGate Pro SMTP 3.5.9)
  with ESMTP-TLS id 4147152; Mon, 19 Aug 2002 17:13:25 +0400
Date: Mon, 19 Aug 2002 17:12:52 +0400 (MSD)
From: malc <malc@pulsesoft.com>
X-X-Sender: malc@home.oyster.ru
To: Oleg <oleg_inconnu@myrealbox.com>
cc: caml-list@inria.fr
Subject: Re: [Caml-list] O'Caml vs C++: a little benchmark
In-Reply-To: <200208181716.NAA10426@hickory.cc.columbia.edu>
Message-ID: <Pine.LNX.4.44.0208191707490.1584-100000@home.oyster.ru>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-caml-list@pauillac.inria.fr
Precedence: bulk

On Sun, 18 Aug 2002, Oleg wrote:

> Hi
> 
> I wrote a few simple benchmarks [1] assessing binaries generated by "ocamlopt 
> -unsafe -noassert" vs binaries generated by "g++-3.2 -O2 -fomit-frame-pointer 
> -pedantic" on P3 Xeon, and the results were quite surprising to me.
> 
> Firstly, I expected iteration over O'Caml lists and integer arrays to be as 
> fast as iteration over std::list and std::vector<int>, respectively. Instead, 
> the benchmark gave me a speed difference of about 10x and 100x in favor of 
> C++ for lists and arrays, respectively.
> 
> Secondly, a little benchmark comparing mutable binary trees of 64 bit floats 
> also showed g++-3.2 to be about an order of magnitude faster.
> 
> What was even more surprising was that O'Caml turned out to be about 10 times 
> faster than C++ for reversing lines in a file. I did not use explicit buffers 
> of any kind in either version, and in C++ program, I used "getline", reading 
> into std::string which should provide about the same level of abstraction and 
> overflow protection as O'Caml string.
> 
> I'm curious as to where these huge differences for these small programs come 
> from.
You made an unlucky choice of summing floats. So what you measure(at least 
in list, array cases) is speed of float boxing.
open Printf;;

let n = int_of_string (Sys.argv.(1));;
let r = int_of_string (Sys.argv.(2));;

type _s = { mutable s : float }
let result =
  let s = { s = 0.0 } in
  let list = Array.to_list (Array.init n (fun i -> i)) in
  let rec f = function [] -> () | x :: xs -> s.s <- s.s +. float x; f xs in
  for i = 0 to pred r do
    f list
  done;
  s.s

Is what you would probably came to, after reading:
http://caml.inria.fr/ocaml/numerical.html
  
I'll try to analyze memory, tree etc, if time permits.

P.S.
lists
1.142
0.510
1.298

(Athlon 1.4G(running at 1G), compiled with -inline 20)

-- 
mailto:malc@pulsesoft.com

-------------------
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