Name: kb87695 Date: 05/17/2001


Steps to reproduce:

1, javac the small testcase foo.java below

public class foo {
    public static void main(String[] args) {

        for(int idx=0; idx < 10000; idx++)
            bar(0x6EEEEEE, 0x5FFFFFFF);

        System.out.println("bar results in " + bar(0x6EEEEEE, 0x5FFFFFFF) );
    }

    static int bar(int a1, int a2) {
        int v1, v2, v3, v4;

        v3 = 0x6FEFEFEF;
        v4 = 0x5FFFFFFF;
        v2 = v3 + 12;
        v1 = v2 + (v4 + 98);
        v2= (a1*v1) * v3;
        v1 = (a2 + v2);
        return v1;
    }
}

2, run the command line
   $ /usr/local/java/jdk1.4/solsparc/bin/java_g -server -Xcomp -XX:CompileOnly=foo.bar foo

3, Error message as below

VM option 'CompileOnly=foo.bar'
#
# HotSpot Virtual Machine Error, assertion failure
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
#
# assert(t->singleton(), "must be a constant")
#
# Error ID: /usr/re/hotspot/hotspot1.4/ws/solsparc/hotspot1.4beta/src/share/vm/opto/phaseX.cpp, 534 [ Patched ]
#
# Problematic Thread: prio=5 tid=0x113c70 nid=0xc runnable
#
Dumping core....
Abort (core dumped)
$

4, Why?
There is a missing check for overflow after calles to mul_ring
and before calls to PhaseValue::makecon in MulNode::Ideal.

It's trying to do the following optimizing tranformation

 (x*con1)*con2 -> x*(con1*con2)

if both con1 and con2 are of singleton type, it wants to make
a constant out of con1*con2. But mul_ring(con1, con2) might overflow
amd return a non-singleton type, i.e. TypeInt::INT, TypeLong::LONG.

5, Java version info
$ java_g -server -version
java version "1.4.0-beta"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.0-beta-b65)
Java HotSpot(TM) Server VM (build 1.4.0-beta-b65-debug, mixed mode)

6. Suggested Fix:
The fix should be to add a check if( t2->singleton() ) before doing the optimization


(Review ID: 124242)
======================================================================