FULL PRODUCT VERSION :
Java HotSpot(TM) Server VM (1.6.0-rc-fastdebug-b90) for windows-x86, built on Jun 30 2006 01:51:43 by "java_re" with unknown MS VC++

FULL OS VERSION :
Microsoft Windows XP [Version 5.1.2600]

A DESCRIPTION OF THE PROBLEM :
The u32*u32=>u64 multiplication from bug 4850194 works only with 2 variables like this:

    private static int DIV3_CONST = (int)0xAAAAAAABL;

    public static int udiv3(int value) {
        return (int)(((value & 0xffffffffL) * (DIV3_CONST & 0xffffffffL)) >>> 33);
    }

but when I declare the DIV3_CONST final it will not use the u32*u32=>u64 marco

I verified this with the following VM options on a fastdebug build:
-server -Xbatch -XX:+PrintCompilation -XX:+PrintInlining -XX:+PrintOptoAssembly


as an alternative to this implement the division by constant that is mentioned in AMD optimisation guide.

THE PROBLEM WAS REPRODUCIBLE WITH -Xint FLAG: No

THE PROBLEM WAS REPRODUCIBLE WITH -server FLAG: Yes

EXPECTED VERSUS ACTUAL BEHAVIOR :
this hotspot macro should also work with one constant
REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
import java.util.Random;

public class X {

    private static int DIV3_CONST = (int)0xAAAAAAABL;

    public static int udiv3(int value) {
        return (int)(((value & 0xffffffffL) * (DIV3_CONST & 0xffffffffL)) >>> 33);
    }

    public static void main(String[] args) {
        Random r = new Random();
        int sum0 = 0;
        for(int i=0 ; i<10000 ; i++) {
            sum0 += Signed.udiv3(r.nextInt());
        }
        System.out.println(sum0);
    }

}
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
do not use final - but this will cause an additional memory access