Re: first test

[Bryce McKinlay <mckinlay at redhat dot com>]

Andrew Haley wrote:

> This isn't a test of the compiler, because it spends almost of of its
> time inside the library calls.  It's really a test of the performance
> of the different Java runtime libraries.
>
> However, I get 3860 ms for the JIT compiler and 4613 ms for gcj.  That
> means that gcj's implementation of strings and/or memory allocation is
> slightly slower.
>  

Actually this test isn't really fair, because a JIT compiler can inline 
the String.valueOf() call and discover that the whole loop is a no-op. 
GCJ could eventually do this too if String.valueOf() doesn't cross a 
binary boundary. Typically, however, programmers don't write code that 
doesn't do anything, so this benchmark is not a valid test of real 
application performance.

Here's a better variant:

public class Loop {

  static final int N = 2500000;
  public static String[] sa = new String[N];

   public static void main(String[] args) {
       long before = System.currentTimeMillis();
       for (int i=0; i < N; i++){
           int a = 27 + 3 * i;
           sa[i] = String.valueOf(a) + "hello";
       }
       long after = System.currentTimeMillis();
       System.out.println("time: " + (after-before));
   }
}

For this case, I get 3203ms for "gcj -O2" and 5013ms for Sun Java 1.5 beta.

Regards

Bryce