> From: Chris Smith [mailto:[EMAIL PROTECTED]]
>
> Artur wrote:
> > I think that JIT compiler should be written for specific VM. Of course
> > it can use the common source for some parts, but decision for which VM
> > it will generate code should be taken at compile time, not at runtime.
> > You propose a way to be able to use same binary JIT library in various
> > VMs. There is no need for it. It is a lot easier to have separate JITs
> > for every VM.
>
> If that's a common viewpoint, then fine. I certainly have no problem with
> using a much simpler interface with the JIT. Unless anyone has a good
> reason to like my previous idea, we can just drop it.
>
> Chris
>
>
Artur is right, especially in that different VMs will probably make special
optimizations that a cross-VM JIT should never make, using internal fields
of jobjects and such.
If it were not for that fact, which pretty much means that the code
generated by the JIT would mostly be VM-specific and therefore unportable, I
would like the idea very much. The more we break these things out, the more
we can help free VMs proliferate.
It is possible to design such an interface, and even make it work well in
the future (expand to JIT'ed HotSpot and such) but it will likely not be
worth the effort. We can share design and optimization techniques between
VMs, at least, even if we can't share code.
--John Keiser
