Peter's slides (was Kaffe's JIT mechanism description)
gback at cs.utah.edu
Wed Nov 11 09:39:41 PST 1998
> Yep. Sorry, I could have said this. But there are some more links (usually in
> white) on the pages. I will add a map when I visit this again - the material is
> an ongoing effort, and comments / contributions are highly appreciated. The
> pics are done with tgif <http://bourbon.cs.umd.edu:8001/tgif/>, in case you
> want to give it a try. Anybody wants to volunteer for a GC picture?
Unfortunately, I wasn't able to attend Peter's talk at the Atlanta
conference, so I don't know whether he clarified the relationship between
what's shown on these slides and what's available.
Basically, the slides don't always show the free, public version of Kaffe,
(although they do for the *most part*.)
For instance, there is no "JDebug" system, the directory structure shown
is slightly different (for instance, there's not kaffevm/jar subdirectory),
there is no msdos-jthreads (the MS-DOS port is part of the custom edition).
When reading the slides, I wasn't sure whether it showed things that
will become part of the public tree (and just haven't been merged),
or whether some mixture of the open and custom edition was being sold
as a GPL'ed system. In any event, I believe it would be beneficial
(and is always a subject of confusion on this list) to point what is
available and under what license, even on these slides.
Also, forgive me for being a bit sarcastic here, the custom edition seems
to have made another breakthrough in JIT technology. To perform 3,000,000
function calls in 205 milliseconds, as claimed in
that is, one function call in 68 nanoseconds on a Pentium/233, is orders
of magnitude faster than what I measured. ;-)
More information about the kaffe