Split Compilation and Code Specialization

Partners

• Dr. Erven Rohou, INRIA
• Prof. Stefano Crespi Reghizzi, Politecnico di Milano

We would like to establish a collaboration among the following parters to stir research on split compilation and its application to code specialization. The investigation will be conducted with a port of the GCC compiler that generates CLI bytecode (aka .NET) for the static part, and with the ILDJIT virtual machine and JIT developed at Politecnico di Milano for the dynamic part.

Background

Split compilation is one direction that attempts to leverage runtime-independent optimization as much as possible, deferring the runtime-dependent part for when we know the execution environment.

Code specialization is a typical optimization that takes advantage of runtime information to achieve good results. The optimizer, though, must decide when the extra work is worth the effort. By statically pre-computing the predicates that will impact the performance, and by embedding them in the bytecode as annotations, we plan to simplify the decision process of the JIT compiler. The optimizer will be able to make a faster and better informed decision.