Re: Speed Impact experiment on GCJ

[Thomas Hallgren <thomas at tada dot se>]

Andrew Haley wrote:
> I tried gcj against IBM's JIT -- still the fastest JIT AFAIAA.
> (Disclaimer: this is an old 32-bit version of IBM's JIT, and a newer
> one may be better.)

I made some similar tests using the HotSpot 1.5.0_06 server JVM on a Fedora Core release 4. 
As you can see, my gcj is slightly older then yours. Don't know how much that matters. 
Results below:

Regards,
Thomas Hallgren


[thhal@localhost gcjtest]$ gcj --version
gcj (GCC) 4.0.0 20050519 (Red Hat 4.0.0-8)
Copyright (C) 2005 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

[thhal@localhost gcjtest]$ java -version -server
java version "1.5.0_06"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_06-b05)
Java HotSpot(TM) Server VM (build 1.5.0_06-b05, mixed mode)

HotSpot server without JIT:

[thhal@localhost gcjtest]$ time java -server -Djava.compiler=none Client task.test servers.d 
result.d

real    0m55.379s
user    0m53.726s
sys     0m1.379s

200.0   7897
300.0   11791
400.0   15660
500.0   19588

HotSpot server JIT:

[thhal@localhost gcjtest]$ time java -server Client task.test servers.d result.d

real    0m0.510s
user    0m0.073s
sys     0m0.393s

200.0   39
300.0   1
400.0   0
500.0   1

REMARK: Obviously, the JIT discovers that the loop is not really a loop sometime
during the first iterations and decides to optimize pretty hard...

gcj at -O0:

[thhal@localhost gcjtest]$ time ./client task.test servers.d result.d

real    0m9.609s
user    0m9.240s
sys     0m0.288s

200.0   1327
300.0   2053
400.0   2662
500.0   3339

gcj at -O3:

[thhal@localhost gcjtest]$ time ./client task.test servers.d result.d

real    0m1.469s
user    0m1.265s
sys     0m0.202s

200.0   184
300.0   276
400.0   346
500.0   442


> I get:
>
> [aph@zorro ~]$ java -version
> java version "1.4.2"
> Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.2)
> Classic VM (build 1.4.2, J2RE 1.4.2 IBM build cxia32142-20041210 (JIT enabled: jitc))
>
> [aph@zorro ~]$ gcj --version
> gcj (GCC) 4.1.0 20060128 (Red Hat 4.1.0-0.17)
> Copyright (C) 2006 Free Software Foundation, Inc.
> This is free software; see the source for copying conditions.  There is NO
> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
>
> IBM without a JIT:
>
> [aph@zorro ~]$ time java -Djava.compiler=none Client task.test servers.d result.d
> Warning: JIT compiler "none" not found. Will use interpreter.
>
> real    1m6.508s
> user    1m6.340s
> sys     0m0.016s
>
> 200.0   9315
> 300.0   13993
> 400.0   18636
> 500.0   23271
>
> IBM with a JIT:
>
> [aph@zorro ~]$ time java Client task.test servers.d result.d
>
> real    0m2.521s
> user    0m2.480s
> sys     0m0.024s
>
> 200.0   334
> 300.0   501
> 400.0   671
> 500.0   835
>
> gcj at -O0:
>
> [aph@zorro ~]$ time ./client task.test servers.d result.d
> time ./client task.test servers.d result.d
>
> real    0m6.982s
> user    0m6.932s
> sys     0m0.024s
>
> 200.0   997
> 300.0   1475
> 400.0   1988
> 500.0   2491
>
> gcj at -O3:
>
> [aph@zorro ~]$ time ./client task.test servers.d result.d
>
> real    0m1.236s
> user    0m1.220s
> sys     0m0.016s
>
> 200.0   167
> 300.0   252
> 400.0   336
> 500.0   425
>
> 32-bit mode doesn't seem to make any difference to us.  gcj at -O3 -m32:
>
> [aph@zorro ~]$ time ./client task.test servers.d result.d
>
> real    0m1.236s
> user    0m1.228s
> sys     0m0.008s
>
> 200.0   167
> 300.0   251
> 400.0   335
> 500.0   418
>
> And, just for laughs, gij:
>
> [aph@zorro ~]$ time gij Client task.test servers.d result.d 
>
> real    0m41.280s
> user    0m41.047s
> sys     0m0.024s
>
> 200.0   5804
> 300.0   8685
> 400.0   11587
> 500.0   14468
>
> Andrew.