Re: JIT pluggability

[Artur Biesiadowski]

Tom Tromey wrote:

> One thing we talked about is the possibility of defining a standard
> pluggable JIT interface, making it easier to mix-and-match JITs and
> interpreters across VMs.
[..snip...]

One quite important thing you have not stressed is gc interface - unless 
gc is supposed to be part of jit, there is a lot of issues to be solved 
(IMHO harder than jit <-> core jvm/classes interface).

For a jumpstart I would suggest looking at orp jit interface.
http://sourceforge.net/projects/orp/
Directory interface, all .h files there.

jit_runtime_support.h - number of jvm function which need to be called 
from _generated_ code

jit_intf.h - jit use these function to discover info about classes it 
needs to compile

jit_export.h - jvm call these methods on jit to request method 
compilation, synchronize gc events and discover stack info

There are gc interface there, but AFAIK, orp never managed to get them 
really pluggable - there was still a number of ifdefs in both jvm and 
jit code depending on type of gc used. In fact, I'm not sure if 
gc/jit/jvm can be separated through abstract interface and do not impose 
any restrictions on implementation.

While orp code uses C for interfaces, I'm not proposing sticking to it 
here - just to scavenge their headers to get a list of functions which 
were needed to implement working, pluggable jit.


Artur