Compiling Kaffe 1.0.5 on Alpha running OSF

Godmar Back gback at
Tue Nov 23 08:57:05 PST 1999

> Thanks for the suggestions about what might be causing the problem with the
> DEC Alpha version of Kaffe.  I would like to see Kaffe up and running on the
> Alpha.  I will try to take another look at the code this weekend.

Edouard told me he thinks the SP_OFFSET is okay.
Another chance maybe that you'll need FP_OFFSET as well.

The third, least attractive possibility is that the setjmp/longjmp
based approach won't work at all on alpha/OSF, which means that
you'll need a custom thread switching routine, similar to the one
for sparc/linux.

Btw, Edouard did some work on trying to get the jit up and running
again, but there's still some problems.  If you're interested in
cooperating on that, you may want to contact him.

> We have a number of multi processor machines in our lab: a 4-processor DEC
> Alpha 4100, four dual-processor Pentium III PCs, and a dual-processor
> pentium II machine. I would like to start pushing some of the issues related
> to performance in Kaffe.  The profiling is one tool to help with that.
> Assuming we can get Kaffe running on the Alpha, we can use Atom and DCPI
> (DEC Continous Profiling Infrastructure) to further analyze the performance
> of Kaffe and to identify areas of improvement.

To support DCPI and DCPI-like systems, I've started including the class; the implementation of some native methods
is missing at this point though.  k.m.JIT allows you to dump information
about where translated methods are located on the heap; this is needed
for DCPI.

	- Godmar

More information about the kaffe mailing list