Performance improvements on the way

Hi all,

I'm glad to announce that several performance improvements are almost ready for actual use.

1) A new implementation for invokeinterface has been implemented that uses precompiled Interface Method Tables (IMT). This is must faster then the previous IMT lookup algorithm. This implementation is now used in the CVS HEAD branch.

2) The new L1A compiler is starting to work. We still need some fixes before it can actually run JNode. This compiler brings several 100K of reduction of generated bootimage code and performs real register allocation. This will deliver a much better performance on both integer & float operations.

I hope that we'll be able to switch to the new L1A compiler in 1 or 2 weeks.


Nice benchmark you got with t

Nice benchmark you got with the new efforts. Good work!

Have you btw. run the test on a Sun J2SE 5.0 RC?


Performance improvements on the way