Comments on: Java VMs Compared

By: Ed Holdsworth

Ed Holdsworth — Thu, 30 Apr 2009 16:14:57 +0000

did you manage to get comparison figures for openjdk?

By: Lizing

Lizing — Sun, 07 Dec 2008 13:38:13 +0000

Hi,

May I ask about what is your ram size that your use in this comparison?

Thank your,
Lizing

By: Robert Schusters Blog » Blog Archive » PhoneME Advanced Foundation (with JIT) at Jalimo

Thu, 18 Sep 2008 10:36:24 +0000

[...] VM. See, this code is still does usefull things for us. Buglabs recently did a comparison of Cacao, JamVM and PhoneME Advanced (interpreted only) on ARM systems. Surprisingly (or not ) [...]

By: Bug Blogger » Blog Archive » Java VMs Compared II

Bug Blogger » Blog Archive » Java VMs Compared II — Wed, 10 Sep 2008 19:08:33 +0000

[...] received great feedback from our previous comparison of Java virtual machines. We acted on all of the comments we received from the Cacao community by tweaking our build [...]

By: gbenson

gbenson — Thu, 04 Sep 2008 15:25:11 +0000

What did you use for your graphs by the way? I like them

By: Christian Thalinger

Christian Thalinger — Thu, 14 Aug 2008 09:47:56 +0000

I just had some spare minutes and did a Scimark2 run on our hard-float ARM box, because I couldn’t get JamVM working on our soft-float board. Have to talk to Robert about this issue. The ARM box does not have a FPA, so the kernel emulates them. Here are the numbers:

JamVM:

SciMark 2.0a

Composite Score: 0.4885741242527907
FFT (1024): 0.29904493861307696
SOR (100×100): 0.6581993778741051
Monte Carlo : 0.29967875915390363
Sparse matmult (N=1000, nz=5000): 0.5820827961957977
LU (100×100): 0.6038647494270701

java.vendor: GNU Classpath
java.version: 1.5.0
os.arch: arm
os.name: Linux
os.version: 2.4.22-iyonix

CACAO:

SciMark 2.0a

Composite Score: 1.0139034019666382
FFT (1024): 0.5376767705902252
SOR (100×100): 1.581337897500241
Monte Carlo : 0.6617709243687945
Sparse matmult (N=1000, nz=5000): 1.273631846838955
LU (100×100): 1.015099570534974

java.vendor: GNU Classpath
java.version: 1.5.0
os.arch: arm
os.name: Linux
os.version: 2.4.22-iyonix

By: Chris Plummer

Chris Plummer — Thu, 07 Aug 2008 06:07:39 +0000

I’m not sure why you had issues with the JIT, since the ARM JIT has always been a stable part of the phoneME Linux/ARM port. However, glad to see you got it working now.

If you have hardware VFP, you should also enable it in the JIT by building with CVM_JIT_USE_FP_HARDWARE=true. Otherwise the JIT will only use GPRs and call out to C functions for all floating point operations, or worse yet, call out to the assembler soft float library.

By: Christian Thalinger

Christian Thalinger — Mon, 04 Aug 2008 14:28:21 +0000

OK, that explains why CACAO is so slow at the floating point benchmarks. But just to be sure, is your toolchain softfloat configured?

By: Kevin Schultz

Kevin Schultz — Mon, 04 Aug 2008 13:21:17 +0000

Christian,

We are using Cacao 0.98 with software floating point enabled. Our CPU is a Freescale mx31 (ARM11), which does have a hardware fpu. We would get a boost out of using that instead. The build does not have -XX:+PrintConfig as a config option, but I will try that, thanks for the tip.

We are using the default OpenEmbedded build scripts rather than tuning them to our hardware, and they must be setup for the most general case. We were quite surprised by Cacao’s performance since we had expected it to be on top so this might explain that discrepancy. Additionally we have gotten big boosts out of phoneME lately now that we have JIT enabled, so I guess a second comparison will be in order.

Thanks,
Kevin Schultz

By: Robert Lougher

Robert Lougher — Sat, 02 Aug 2008 05:44:01 +0000

From the documentation the bug uses a ARM1136JF-S core, which has a VFP coprocessor… I guess CACAO is the OpenEmbedded version. Ask Mr. Schuster. He’s just packaged PhoneME on OpenEmbedded so I guess it’s a moot point anyway…