[Caml-list] novice puzzled by speed tests

[dolfi@zkm.de]

Hi everybody,

I'm new to the list, and deeply impressed by the stability and speed of
OCaml, as well as its functional nature. My best congratulations to the
authors!

Toying around with 3.07, I found that ocamlopt.opt -unsafe (on Mandrake
9.1, Pentium 4, 2.4 GHz) actually produces slower code than ocamlopt.opt.
The program in question is a prime generator:

let count = 400000

let ptab =
  let ptab = Array.create count 0 in
  ( ptab.(0)<- 2; ptab.(1) <- 3; ptab.(2) <- 5; ptab );;

let rec loop nr toggle psize =
(   let max = truncate (sqrt (float nr)) in
    let rec floop i =
      if ptab.(i)>max then
        ( ptab.(psize) <- nr;
          loop (nr+toggle) (6-toggle) (succ psize) )
      else if nr mod ptab.(i) = 0 then
        loop (nr+toggle) (6-toggle) psize
      else
        floop (succ i)
    in if psize<count then floop 2 else ()
) in loop 5 2 3;

Printf.printf "prime %d: %d\n" count ptab.(pred count);;

On my box, the corresponding C program (gcc -O3) is slightly slower than
the ocamlopt.opt compiled O'Caml program, but about 25-30% faster than the
-unsafe one:

#include <stdio.h>
#include <math.h>

#define how_many 400000

int main()
{   unsigned int nr = 5, toggle = 2, max, primes_size = 3, i;
    unsigned int primes[how_many];

    primes[0] = 2; primes[1] = 3; primes[2] = 5;
  loop:
    nr += toggle; toggle = 6 - toggle;
    max = sqrt(nr);
    for (i = 2; primes[i] <= max; ++i)
        if (!(nr % primes[i])) goto loop;
    primes[primes_size++] = nr;
    if (primes_size < how_many) goto loop;

    printf("%i\n", primes[how_many - 1]);
    return 0;
}

Of course it's good that range checking increases the speed of programs,
but, being a long-time C user, I'm a little bit puzzled by miracles like
this. I suspected that the sense of the -unsafe flag was inverted, but it
isn't: the -unsafe program dies with SEGV when I deliberately introduce a
range overflow, the safe one gets an exception.

Till soon, Dolfi




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