[freenet-dev] Re: Portability issues: integer ++ atomicity?

[Michael Schierl]

Toad schrieb:
> 
> Is ++'ing an integer always atomic in java, assuming the var is declared
> volatile? AFAICS, x86 would be atomic... but anything RISC would involve
> a separate load and store, right?

it is not - even on x86. 

Looked though my java files and found the one i used to test this.

mihi
-------------- next part --------------
class ThreadDemo implements Runnable {
    private Thread t;
    private volatile int local;
    private static volatile int global;
    public static final int COUNT = 2;
    public static int HOWFAR = 10000000;
    public static int TESTNUM = 10;
    public static Object o = new Object();

    public ThreadDemo () {
        (t=new Thread(this)).start();
    }

    public void run () {
        int max = (int) (HOWFAR * Math.random());
        for (int i=0;i < max; i++) {
            local++;
            global++;
        }
    }

    public static void main(String[] args) {    
        for(int i = 0; i<TESTNUM; i++) 
            dotest();
    }

    public static void dotest () {
        ThreadDemo[] tda = new ThreadDemo[COUNT];
        global=0;
        for (int i=0; i<tda.length; i++) {
            tda[i] = new ThreadDemo();
        }
        long sum = 0;
        for (int i=0; i<tda.length; i++) {
            try {
                tda[i].t.join();
            } catch (InterruptedException e) {System.exit(0);}
            sum += tda[i].local;
        }
        System.out.println
            ("Lokal: "+sum+", Global: "+global+ ", also "+
             (sum==global ? "okay" : "Fehler: "+ (sum-global)));
    }
}