[kaffe] interpreter performance

[Dalibor Topic]

On Tue, Jan 17, 2006 at 06:11:20PM +0100, Lorenz Witte wrote:
> I've been testing interpreter performance of Kaffe and JamVM on x86 
> using the Embedded Caffeine Mark test and I found JamVM to run about 10 
> times as fast as Kaffe. Are these numbers reflecting reality or is there 
> another explanation?

JamVM has a modern, highly optimised interpreter, and does not have a
jit (or need one :). Kaffe, otoh, has a pretty simple interpreter,
that's supposed to help you port Kaffe easily to your platform, so that
you can get on with making the jit work quickly. Given the two different
design goals, JamVM's interpreter is bound to be faster than Kaffe's
interpreter, and Rob's pretty good at sqeezing performance out of it :)

If you are interested in improving Kaffe's interpreter, Cacao now
includes a highly optimised vmgen-generated interpreter core, that could
be merged into Kaffe if someone volunteers. Christian will be talking
about his work on integrating it with Cacao at FOSDEM, and I'd like to
see it turned into a component that other GNU Classpath VMs can plug in,
as that would allow people to focus on opitimising the interpreter
generator, rather than each VM reinventing the tricks in their
interpreter engines.

cheers,
dalibor topic

> 
> Lorenz
> 
> 
> _______________________________________________
> kaffe mailing list
> kaffe at kaffe.org
> http://kaffe.org/cgi-bin/mailman/listinfo/kaffe