Re: Performance comparison of Java (and .NET) runtimes

[shudo at computer dot org]

Hi Bryce,

> >  Performance Comparison of Java/.NET Runtimes (Oct 2004)
> >  http://www.shudo.net/jit/perf/

> Cool, thanks for posting that. It would be good to understand why GCJ
> does so poorly on some of the SPEC JVM scores. Did you ever try it with
> -ffast-math and/or -march=pentium4? (also, perhaps -fno-bounds-check,
> but I guess that is cheating)

Now I tried those options and saw some amount of improvement.
The score of _222_mpegaudio improved 15% and the total score improved 4%.

[Only -O2]

  _227_mtrt       122
  _202_jess       87.0
  _201_compress   187
  _209_db         29.0
  _222_mpegaudio  375
  _228_jack       56.1
  _213_javac      103

[-O2 -march=pentium4 -ffast-math]

  _227_mtrt       125
  _202_jess       90.0
  _201_compress   190
  _209_db         29.3
  _222_mpegaudio  430
  _228_jack       56.7
  _213_javac      107

The version of the gcj I used is as follows:

  % gcj -v
  ...
  gcc version 3.4.2 20041017 (Red Hat 3.4.2-6)

  % rpm -q gcc-java
  gcc-java-3.4.2-6


  Kazuyuki Shudo	shudo@computer.org	http://www.shudo.net/