Performance of ZipPy when running richards.py

[Wei Zhang]

Hi all,

ZipPy has migrated to the new inlining API for a while.
I also rewrote Python object accesses and method calls to use a polymorphic
inline caching dispatch.

I've been struggling with the performance of ZipPy when running the
richards benchmark and was able to get it to 'very' close to PyPy (2.2.1).

richards:
ZipPy: 2.995 sec
PyPy: 2.254 sec

I'm puzzled that why ZipPy is still a bit slower than PyPy after object
accesses and calls are this optimized.

I have extracted the hottest part (that I believe) of the benchmark to a
reasonably small micro benchmark and anybody can run it with the following
command in a checked out ZipPy repo:
./mx.sh python graal/edu.uci.python.benchmark/src/micro/boolean-logic.py

Can someone from the Truffle team please have a look at that and maybe give
me some advice on what I can improve?

Special thanks to Andreas Woess for answering my questions on skype.

Thanks,

/Wei