Re: DaCapo benchmarks

[Andrew Haley <aph at redhat dot com>]

I found out why gcj is so slow on ps, and it seems to be that
ps is throwing millions of exceptions.  It boils down to this method:

    /**
     * Returns the value associated with the key term.
     * @param key Key term.
     * @return PSObject No such value -> null, else -> associated value.
     */
    public PSObject getValueOf( PSObject key ) throws PSObjectException
    {
	/* attempting to get value from an empty dictionary */
	if( ((ArrayList)value).isEmpty() )
	{
	    throw new PSObjectException( "empty dictionary" );
	}

	/* Iterates through the arraylist, searching for the pair with the
	   given key term. */
	PSObject returnObject = null;
	for( int i = 0; i < ((ArrayList)value).size(); ++i )
	{
	    if( ((Pair)((ArrayList)value).get( i )).isKey( key ) )
	    {
		returnObject = ((Pair)((ArrayList)value).get( i )).getValue();
		break;
	    }
	}

	/* Can't find the key at all. */
	if( returnObject == null )
	{
	    throw new PSObjectException( "can't find key" );
	}
	return returnObject;
    }

If you look at its spec, getValueOf() is supposed to return null if a
value is not found.  However, it does not do that: it throws an
exception instead.

The gcc exception processing mechanism is designed to be very fast for
the "never thrown" case, but it is slower when exceptions are thrown.
I suppose gcj could change to optimize for things like this, but IMO
it would not be such a good idea for "reasonable" applications.

I'm very tempted to fix getValueOf() so that it does what its
specification requires, but that would invalidate the benchmark.

Andrew.