[groovy-dev] Groovy JIT and Partial Sums Benchmark

Hello all,

I've been writing a JIT for Groovy in my GSOC period, and I've got the
alpha code quality with some figures to show.
When a JIT technique, you don't need to worry about MOP representation
in bytecode as long as the default metaclass is not changed.
I use JVMTI to make this happen (I'm lucky enough to find the way to
avoid Java 6 specific feature and able to implement it for Java 5).

I'm confident that we are now able to get Java-near speed for Groovy 

Before going to the figures, I'd like to thank you:
  - Guillaume to let me in GSOC program this year,
  - Alex T. for making callsite fast enough to make me to find an
alternative way - and look back to my GJIT project,
  - Jochen for his questions and thoughts of how to remove
"box/unbox"-ing pairs.

This partial sums benchmark is from shootout. I've selected it to be
the first one because the ease of implementation.
As I mentioned, the JIT quality is still alpha, but at least I've got
a good enough code structure to patch (and patch).
My current goal is to beat all shootout benchmarks at the moment, and
I hope to finally have some more figures to show.

- PartialSumsJ is Java implementation of partial sums, taken from shootout.
- partial_sums.groovy is untyped version, taken from shootout.
- alioth/PartialSums.groovy is typed version, modified from partial_sums.groovy
All results use Java 1.6_06 with -server option.