java compiler optimization

Question

Is the Java compiler smart enough to optimize loop below, by extracting the

Double average = new Double( totalTime / callCount ); 

out of the for loop?

public double computeSD( Set values, int callCount, long totalTime ) {
  double diffs = 0.0d; 
  for( Iterator i=values.iterator(); i.hasNext(); ) {
    double value = ( ( Double )i.next() ).doubleValue(); 
    Double average = new Double( totalTime / callCount ); 
    diffs += ( value – average.doubleValue() ) * ( value – average.doubleValue() );
  } 
  double variance = diffs / callCount;
  return Math.sqrt( variance );
}

Answer 1

Java will not and can not extract it from the loop. Any use of the 'new' keyword will always result in a new object being created. You would be better off using Double.valueOf()

See the javadoc for Double.valueOf(double):

"Returns a Double instance representing the specified double value. If a new Double instance is not required, this method should generally be used in preference to the constructor Double(double), as this method is likely to yield significantly better space and time performance by caching frequently requested values."

If you used this method it would return the same object every time, thus reducing the number of objects created and increasing performance.

However, using valueOf is still not the answer for you!

valueOf is still a method call, and method calls do not get optimized away. It will call valueOf every iteration of the loop. Look through your method and count the method calls. Right now it is 6, including hasNext and new Double, which is similar to a method call. These will all happen every time, and no java optimization will change that. You are better off refactoring to remove as many method calls as possible from the loop.

Answer 2

Nothing prevents the bytecode compiler (java->bytecode) from performing optimizations. When I worked at Symantec, and they did a Java IDE, the compiler write did look into putting some optimizations into our compiler but said nobody (in the outside world) seemed to be interested and the focus was on the Just In Time (JIT) compiler that is roughly the same as HotSpot in modern Sun VMs.

There is nothing that prevents a bytecode compiler from performing optimizations, but I am not aware of any that do so. There is huge focus on runtime optimizations, but those are pretty much hidden at runtime.

So, the source->bytecode compiler probably does not optimize it but the VM probably does. If you are on something like an Android then it probably performs no runtime optimization.

Answer 3

This may seem like an obvious optimization at first, but I don't think so, since it involves object instantiation. Granted, it's an instantiation of an immutable primitive box type, but that still doesn't guarantee that there's no side effect.

I don't think any current compiler can optimize this. For this to be optimized, the compiler must be told that some classes have special properties (which can be a dangerous proposition given that things may change in the future). That is, the compiler must be told specifics of the API. This can not be optimized at the language-level alone.

If you use double, however, it's much more likely to be optimized (e.g. using the loop-invariant code motion technique).

Answer 4

If you really want to be sure, the answers to this question tell you how to see the native code that the JIT compiler produces.

Answer 5

Not really. The compiler just writes out byte-code. If anything optimized the code, it'd be the Java Virtual machine and that probably depends on the platform, implementation and execution conditions...