Kaffe code optimisation
tim at transvirtual.com
Fri Sep 10 10:23:33 PDT 1999
You might want to start from the new JIT3 in the CVS tree - it does better
instruction selection (and register allocation come to that) than the
older JIT. Mind you, currently I've not checked in the SPARC port of JIT3
(because my SPARC machine is broken right now) but I can check this in
'untested' so you could use it as a starting point.
On Fri, 10 Sep 1999, Veeral Shah wrote:
> I am trying to implement a new register allocation algorithm (run time
> on the native code generated by kaffe on ultra-sparc processor..
> My problem is that the raw native code which kaffe generates is too
> unoptimised for
> the above algorithm to give better run time result ..I even dont rule
> out the possibility of
> first optmising the native code (using standard algorithms for 1)copy
> propagation 2)dead code elimination ) in an extra parse .. But i cant
> figure out the effort (in man hours) that will be involved in that phase
> as my aim is to check only the register allocation algo performance on
> run time code..
> If any developer/researcher who has already tried doing optimisation
> on the code could he
> guide me / suggest any idea by which i can reduce the effort involved in
> optimisation phase ..
> Any suggestion or hint will be a great help..
> Thanks a lot in advance..
> Veeral Shah
More information about the kaffe