This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
Re: GCJ Benchmarkes
- From: Mathieu Lacage <Mathieu dot Lacage at sophia dot inria dot fr>
- To: Marco Trudel <mtrudel at gmx dot ch>
- Cc: GCJ <java at gcc dot gnu dot org>
- Date: Tue, 28 Nov 2006 08:23:19 +0100
- Subject: Re: GCJ Benchmarkes
- References: <456ACC61.4000800@gmx.ch>
On Mon, 2006-11-27 at 12:30 +0100, Marco Trudel wrote:
> Hey Guys
>
> Did anyone benchmark GCJ? I'm searching for some good tests to compare
> GCJ to a 1.5 Sun JVM. So far I found this benchmark:
> http://www.cs.cmu.edu/~jch/java/linpack.html
> It will run about 1.2 times faster on GCJ compiled applications than
> with a Sun JVM.
I have done a lot of benchmark on an application of mine. It took me a
good few days to make the gcj version faster than the sun hotspot
version, despite gcj using non-conformant floating point operations and
the application being heavily floating-point driven. In the end, I got
the gcj version 1.5 times faster.
In my case, the killer issues were type checks and calls through virtual
methods which are hard to inline correctly across compilation units for
gcj. The SUN hotspot, on the other hand, can do this trivially so, this
is where I saw the biggest difference.
This was 1 year ago and I did not write any java code since then so,
take this with a grain of salt.
regards,
Mathieu
--