benchmark result

Mathieu Lacage Mathieu.Lacage@sophia.inria.fr
Tue Dec 7 09:58:00 GMT 2004


On Mon, 2004-12-06 at 16:07 +0100, Mathieu Lacage wrote:
> On Mon, 2004-12-06 at 13:56 +0000, Andrew Haley wrote:
> > We could improve the performance of these collection classes with
> > little work.  What is required is someone to study the code, do some
> > profiling, and fix things.
> 
> I will look into this when I find some time.

I tried to replace my iterators by getters on the critical paths of the
code. The patch itself is rather simple and I found the results
interesting...

jdk 1.4.2 before:
real 7.88
user 7.64
sys 0.08

jdk 1.4.2 after:
real 10.12
user 9.86
sys 0.05

jdk 1.5.0 before:
real 11.27
user 10.89
sys 0.11

jdk 1.4.2 after:
real 14.10
user 13.78
sys 0.06

gcc 4.0.0 has been built with USE_MARK_BYTES, code has been compiled
with -ffast-math -mtune=pentium4 -O2

gcc 4.0.0 before:
real 8.65
user 8.41
sys 0.05

gcc 4.0.0 after:
real 5.75
user 5.20
sys 0.04

So, now, I get the most performance out of gcj by far when I optimize
the code for it but I must say I am rather unhappy with the idea of de-
optimizing my code for the sun jdks...

regards,
Mathieu
-- 



More information about the Java mailing list