[PATCH] speed up java.util.Random

[Paolo Bonzini]

I saw a PR some time ago about the slowness of java.util.Random in GNU
classpath.  I cannot remember and cannot find the PR number, and
actually I cannot reproduce this slowness when comparing the classpath
implementation against the Sun implementation:

Sun JDK 1.4.2         4586  6038  6216  6254  6875  6952    6153.50
current classpath     3398  3586  3996  4169  4228  4962    4056.50

The huge variance is because I am compiling and running the benchmarks 
with Sun's JITting virtual machine.

Still, inlining the next(int) method, while keeping the same result and
the synchronized behavior of java.util.Random, I could improve
performance by ~20%:

current classpath     3398  3586  3996  4169  4228  4962    4056.50
patched classpath     2585  2618  3465  3538  3606  3807    3269.83

I don't know how this can be contributed to GNU classpath.  I attach a
patch with a changelog, and the tests I ran to validate my
implementation and to compare the relative speeds.

Paolo

2005-11-24  Paolo Bonzini  <bonzini@gnu.org>

         * java/util/Random.java: Rewrite, inlining the next(int) method
         and synchronization blocks into the callers.


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: RandomTest.java
URL: <http://gcc.gnu.org/pipermail/java-patches/attachments/20051125/bff32c7d/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: RandomSpeed.java
URL: <http://gcc.gnu.org/pipermail/java-patches/attachments/20051125/bff32c7d/attachment-0001.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: faster-random.patch
URL: <http://gcc.gnu.org/pipermail/java-patches/attachments/20051125/bff32c7d/attachment-0002.ksh>