Section: New Results

Around processor virtualization

Participants : François Bodin, Christophe Levointurier, Sylvain Leroy, Erven Rohou, Thierry Lafage, André Seznec.

The usage of the Java language has been generalized in the past few years. Applications are now very large and are deployed on many different platforms, since they are highly portable. However ensuring code quality maintenance and code security on those applications is challenging. To address these issues, we are defining a refactoring platform for Java. Java has popularized the distribution of software through bytecodes. Functional portability is the main argument for such a usage of bytecodes. With the new diversity of multicore platforms, functional, but also performance portability will become the major issue in the next 10 years. We have initiated a research effort to efficiently compile towards bytecodes.

Analysis and transformation of Java codes

Participants : François Bodin, Sylvain Leroy, Christophe Levointurier.

The growing size of web applications makes it more and more difficult to keep the code secure with a high quality level. This is a challenge to developers. Code refactoring is one of the techniques to address this issue. However it must be automated to come at low cost and high productivity. The Serenitec (Security analysis and Refactoring ENvironment for Internet TEChnology) platform provides a flexible set of tools to implement Java source code analysis and refactoring.

Refactoring functions are expressed using sets of scripts written in the Java language. The complex analysis or refactoring operations are broken down into several simple scripts that are pasted into a task graph similar to Kahn's networks. For achieving fast processing of large applications (millions of lines of code) the scripts/tasks can then be executed in parallel if they deal with different parts of the code. Analysis and refactoring operations are gathered in reference sets. Each reference set aims to deal with a standard of coding rules. Serenitec offers static program analysis to ensure the correctness of source code transformations. Current implementation achieves high execution speed. For instance, interprocedural symbol renaming (variable, classes and methods) across large applications can be performed in a few minutes using a standard workstation. Serenitec can also be used in continuous integration engines such as Hudson [41] .

Split vectorization

Participants : Erven Rohou, Thierry Lafage, André Seznec.

We attempt to reconcile two apparently contradictory trends of computing systems. On the one hand, hardware heterogeneity favors the adoption of bytecode format and late, just-in-time code generation. On the other hand, exploitation of hardware features, in particular SIMD extensions, is key to obtaining the required performance.

We want to show that vectorized bytecode is a viable approach, that can yield the expected speedups in the presence of SIMD instructions, and a minor penalty in its absence. We also analyze the interaction between the static and the JIT compilers and we devise suggestions for a good performance of vectorized bytecode.

This research is done within the framework of the HIPEAC2 network in collaboration with Albert Cohen (INRIA Alchemy), Sami Yehia (Thalès Research), Ayal Zaks and Dorit Nuzman (IBM research).

Application of Split Compilation to Code Specialization

Participant : Erven Rohou.

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. We plan to statically pre-compute the predicates that will impact the performance, and to embed them in the bytecode as annotations, in order to simplify the decision process of the JIT compiler. The optimizer will be able to make a faster and better informed decision.

This research is done in collaboration with Prof. Stefano Crespi Reghizzi from Politecnico di Milano.