From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: weis
Received: (from weis@localhost) by pauillac.inria.fr (8.6.10/8.6.6) id QAA07652 for caml-redistribution; Tue, 12 Mar 1996 16:38:01 +0100
Received: from concorde.inria.fr (concorde.inria.fr [192.93.2.39]) by pauillac.inria.fr (8.6.10/8.6.6) with ESMTP id GAA29064 for <caml-list@pauillac.inria.fr>; Tue, 12 Mar 1996 06:29:48 +0100
Received: from simon.cs.cornell.edu (SIMON.CS.CORNELL.EDU [128.84.154.10]) by concorde.inria.fr (8.7.1/8.7.1) with SMTP id GAA22308 for <caml-list@pauillac.inria.fr>; Tue, 12 Mar 1996 06:29:53 +0100 (MET)
Received: from cloyd.cs.cornell.edu (CLOYD.CS.CORNELL.EDU [128.84.227.15]) by simon.cs.cornell.edu (8.6.10/R1.4) with ESMTP id AAA21105 for <caml-list@pauillac.inria.fr>; Tue, 12 Mar 1996 00:29:46 -0500
Received: from verdandi.cs.cornell.edu (VERDANDI.CS.CORNELL.EDU [128.84.254.202]) by cloyd.cs.cornell.edu (8.6.10/M1.8) with ESMTP id AAA02586; Tue, 12 Mar 1996 00:29:44 -0500
Received: (hayden@localhost) by verdandi.cs.cornell.edu (8.6.10/C1.3) id AAA02769; Tue, 12 Mar 1996 00:29:41 -0500
Date: Tue, 12 Mar 1996 00:29:41 -0500
Message-Id: <199603120529.AAA02769@verdandi.cs.cornell.edu>
From: Mark Hayden <hayden@cs.cornell.edu>
To: caml-list@pauillac.inria.fr
cc: hayden@cs.cornell.edu
Subject: Caml Light performance tests?
Sender: weis



Hi,
  I'm fine-tuning a program that compiles with both Caml Light and Caml
Special Light.  In doing so, I've made lots of guesses about the relative
performance of different coding styles.  I'm thinking of putting together
some performance tests to eliminate my guesswork.  As you'll see, I'm not
interested in real application benchmarks, but just the relative costs of,
for instance, iterating over a vector with "do_vect" verses an explicit
"for" loop.  I have several questions for the Caml Light community:

  * Has anyone else already made such tests?  (If so, what
    were the results and is the code available?)

  * What are the issues for the cases below?

  * Are there suggestions for other interesting tests?
  
thanks,
  Mark



options:
  Caml Light, CSL-byte, CSL-native
  safe & unsafe
  architecture (I'm interesting in sparc 20s)

1.
  let f = fun x -> ... in
  ...
  
  do_vect f v

    VS

  do_vect (fun x -> ... ) v ;

    VS 

  for i = 0 to pred (vect_length v) do
    ...
  done 

2.

  do_list (fun x -> ... ) l

    VS 

  let rec loop = function
    | [] -> ()
    | hd::tl -> (
	... ; 
	loop tl
      )
  in loop l

3. 

  let g = f x in
  for i = 1 to 1000 do
    g y
  done

    VS

  for i = 1 to 1000 do
    f x y
  done

4.

  let f () =
    if ... then None
    else Some ...
  in

  for i = 1 to 1000 do
    match f () with
    | None -> ()
    | Some x -> ...
  done

    VS

  let f g =
    if ... then g ...
  in

  for i = 1 to 1000 do
    f (fun x -> ...)
  done


4. cost of calling identity function

5. cost of an additional argument to a function

6. overhead of calling an external C function

7. cost of a let binding (is there any?)