Re: Benchmarks (who has the fastest free VM)

[shudo]

Hi Dalibor,

> > I agree most JVMs have not implemented the FP-strict semantics except
> > JDK 1.4 and later, BulletTrain (an AOT compiler) and shuJIT.
>
> speaking out of speculative interest, is shuJIT licensed in a GPL compatible
> way? could one link kaffe with it?

ShuJIT is distributed under the terms of GNU LGPL.  On that point, we
all can link it with Kaffe.  But it does not work with Kaffe for
several reason.  Kaffe does not provide the Sun's old JIT interface
and the JIT make use of internal structures of Sun's Classic VM as
well as the JIT interface.

> i think shuJIT is quite interesting, since it uses sun's JIT deprecated
> interface. I'm not aware of any free JVM implementing that interface.

OpenJIT (http://www.openjit.org/) also uses the JIT interface, though
I'm not sure you look it as 'free'.

> I'm wondering if a free software implementation of a glue interface
> to java jitters would be desirable.

I hope we can design a comfortable and efficient JIT interface and
come to an agreement to adopt it.  I think it is possible that GCJ's
interpreter (gij) turns into having a JIT interface and other JVMs
adopt it.

Designing a clean and efficient JIT interface has been a research
topic.  ORP provides one of such interfaces and Open Virtual Machine
project (http://www.ovmj.org/) led by Jan Vitek had a plan to develop
a JIT interface and uses OpenJIT.

My concern for developing a JIT interface is a method of access to
internal structure of a JVM.  For example, each JVM has its own memory
layout of an object, which can be a characteristic part of the JVM.  A
standard and interoperable interface can be designed but such an
interface is usually not so efficient like JNI.

  Kazuyuki Shudo        address@hidden  http://www.shudo.net/