Strange double calculation result with JDK1.4.2

Performance Tuning
#1

Hi,

I’ve written a small benchmark to test the processing power of the new Sharp Zaurus PDA models.http://www.alterna.tv/zjb/Zjb.java

Then I wanted to compare the result with my workstation, with JDK 1.4.2 Beta. And I was really surprised to discover that the Pi calculation was giving a wrong result: Pi=3.1413934230804443!

This only happens when running the whole program AND with JDK1.4.2.
I had no problems when:

  • using other JDK: 1.4.1, 1.4.1_01, 1.3.1_01, 1.1.8
  • using other platforms: PC, Zaurus
  • extracting the Pi calculation fragment, with/without previous code…

So it must be:
0) A bug in my code, but I can’t pinpoint it.

  1. A problem in my workstation configuration. Apart from setting up JDK1.4.2, nothing new has happened recently.
  2. A bug in the platform/OS. Could a new Win2K hotfix have broken a fix for a hardware problem, and Sun coincidentaly has removed the support from the JVM :stuck_out_tongue: My platform is a Win2K SP3 2xPII 350MHz.
  3. A normal bug encountered in Beta software. Never play with Beta software keep saying my mother :wink:
  4. An alien cause not listed in software development courses…

Could someone with a 1.4.2 JDK setup try my code and check the Pi value?
There’s no need reporting the problem to Sun if it’s not point 2) or 3).

As a bonus, you’ll can compare the performance of your workstation with a Zaurus SL-5500 PDA running Jeode http://www.alterna.tv/zjb/zjb-SL-5500.png or Sun Personal Profile http://www.alterna.tv/zjb/zjb-SL-5500-Sun.png.

#2

Check the strictfp keyword:
http://java.sun.com/docs/books/jls/second_edition/html/expressions.doc.html#249198
I’m not sure if it will fix PI for you but it could explain your problem and how it could be legal.

#3

Hi leknor,

Thanks. I’ve forgotten to say that I had already tried the strictfp modifier and it does not help: I got the same result on my workstation…

That’s really strange, because I can switch the JVM in Eclipse (or on the command line) at will, and I get a different result only on that test! And on the Zaurus with FP emulator, I get it right!

But I have no another PC where to try another JDK1.4.2 setup and check if it’s a configuration problem… :stuck_out_tongue:

#4

OK, I’ve found another PC where to setup JRE 1.4.2Beta. The problem still occurs and seems to be related to the new JVM (Sun Bugs #4843257 & 4828693).

Conclusion: don’t use the 1.4.2Beta with intensive float calculations!