The fix for 5107650 accidentally removed the identity for x | 0. Previously this was handled by AddNode::Identity but when OrINode::Identity was introduced it didn't call it's super so now x | 0 is no longer being optimized. This was noticed in spec/jbb/infra/Util/DisplayScreen.putInt. The code there uses a bitwise or instead of a logical or for a range check, but this shouldn't result in any different code, apart from any reordering allowed by different code shapes. here's a test case with both forms.

public class or {
    static public int v = 0;
    static public int l = 10;
    static public int e = 0;
    
    public static void main(String[] args) {
        for (int i = 0; i < 100000; i++) {
            logic_or(i % 10);
        }
        for (int i = 0; i < 100000; i++) {
            bitwise_or(i % 10);
        }
    }
    public static void bitwise_or(int x) {
        if ((x < 0) | x >= l) {
            throw new InternalError("out of range " + x);
        } else {
            e++;
        }
    }
    public static void logic_or(int x) {
        if ((x < 0) || x >= l) {
            throw new InternalError("out of range " + x);
        } else {
            e++;
        }
    }
}

OrLNode has the same problem.
###@###.### 2005-03-14 20:47:08 GMT