Peter's slides (was Kaffe's JIT mechanism description)

peter at peter at
Fri Nov 13 20:20:28 PST 1998

Godmar: It wasn't a modified VM (though, the custom version), but I used just
builtin types, static methods, no gc, and added array bounds checking to the C
code (which wasn't optimized).

Enno: If you see one of these sarcastic mumblers again, tell them to run the
call test with optimized gcc code, and they will get a big surprise: even
slower! The rest is a bit faster (because of the loops), but that probably is
just a matter for a better register allocation (and/or peephole optimization)
and better float code (as I mentioned). The point is: if you don't compare
apples and peaches, we already *have* comparable speed with JIT-Java and C. And
with the WAT (AOT), or second stage JIT, there is no reason why we shouldn't
have the *same* speed. Of course, that doesn't mean we can already lean back
(remember the last JIT slide about the "optimal" JIT?).

As to the talks: the goal was to show the nature of an OpenSource project,
where you don't talk about announcements, upcoming betas , "cheap" support
programs and all that ad stuff. It's a healthy whitebox development, where
implementation and technical discussion matters, and *we* are the people who can
do that. As I said several times, it wasn't about understanding the bits&bytes,
but about understanding what level of detail, "openess" and, yes, controversial
tech discussions are involved in such an OpenSource project. Of course, we also
have to intro Kaffe, and to do the demos (so that it isn't too dry). If somebody
has a better idea of how to pack that into 45 minutes, please don't hesitate to
let me know (even though I got very positive feedback). Let me put this
straight: this is a combined effort, don't lean back and wait for us to do the
show, to "sell" something that we all publish under GPL. If you don't do that
already - stand up, contribute, and make things better.

-- Peter

More information about the kaffe mailing list