Re: Announcing JGNAT public version 1.0p

[hugin777@my-deja.com]

In article <38ECB0CC.3B2941E2@earthlink.net>,
  Charles Hixson <charleshixsn@earthlink.net> wrote:
> [..] This will let me do the screen designs in Java, where there are
tools around
> and portable libraries that will (almost always) already be installed,
and the
> drive them with an Ada main program that calls the screens to get
information.

I wonder why you would do that ?

AFAICS the only reason could be low-level system interaction (which is
not portable:-).

The problem with Java is the amount of memory consumed by the JVM, not
the speed anymore (again AFAICS).

Try this in IBM's JDK 1.1.8:

public class Primes {

    public static void main(String args[]) {
        final int MAX_PRIME = Integer.parseInt(args[0]);

        long start = System.currentTimeMillis();
        boolean[] isPrime = new boolean[MAX_PRIME];
        for (int i = 2; i < MAX_PRIME; i++) isPrime[i] = true;
        for (int i = 2; i < MAX_PRIME; i++) {
            if (isPrime[i]) markMultiplum(i, isPrime);
        }
        long end = System.currentTimeMillis();

        for (int i = 2; i < MAX_PRIME; i++) {
            if (isPrime[i]) System.out.print(i + ", ");
        }
        System.out.print("duration: " + ((end-start)/1000.) + "
seconds.");
    }

    // This better get inlined !!
    // (Oh, well, the compiler won't read this, will it ? :-)
    private static void markMultiplum(int divisor, boolean[] isPrime) {
        for (int i = divisor+1; i < isPrime.length; i++) {
            if (isPrime[i] && i % divisor == 0) isPrime[i] = false;
        }
    }
}

<-- cut

Note that "markMultiplum" actually get inlined by IBM's JIT !

Then try this in GNAT 3.12:

with Ada.text_io; use Ada.text_io;
with ada.command_Line; use ada.Command_Line;
with Ada.Calendar; use Ada.Calendar;

procedure Primes is
    type BoolArray is array (Integer range <>) of Boolean;
    --------
    procedure MarkMultiplum (Divisor  : in     Integer;
                             Is_Prime : in out BoolArray) is
    begin
        for i in Divisor+1 .. Is_Prime'Last loop
            if i mod Divisor = 0 then Is_Prime (i) := false; end if;
        end loop;
    end MarkMultiplum;
    pragma Inline (MarkMultiplum);
    --------
    arg      : String               := Argument (1);
    Max      : Integer              := Integer'Value (arg);
    Is_Prime : BoolArray (2 .. Max) := (others => true);

    start, finished : Time;
    tm : Duration;
begin
    start := Clock;
    for n in Is_Prime'Range loop
        if Is_Prime(n) then
            MarkMultiplum (Divisor => n, Is_Prime => Is_Prime);
        end if;
    end loop;
    finished := Clock;
    tm := finished - start;

    for I in Is_Prime'Range loop
        if Is_Prime(i) then Put (i'Img & ","); end if;
    end loop;

    Put_Line (tm'Img & " seconds.");
end Primes;

<-- cut

On my machine (with gcc -O3) Java is 3 times faster !!  (Due to Pentium
optimizations, I guess)

Neat 8-)

YMMV of course. These kind of "pseudo benchmarks" are of course not
worth much - but it gets my point through, anyway :-)


PS: I haven't tried JGNAT yet, unfortunately. But I hope the Java
version from the Ada source will be as fast as the one from the Java
source...

Regards,
    Jens Jakob Jensen, Denmark.


Sent via Deja.com http://www.deja.com/
Before you buy.