Section: Overall Objectives

Quick view of Compsys II Achievements and directions for Compsys III

The main achievements of Compsys II were:

• the great success of the collaboration with STMicroelectronics with many deep results on SSA (Static Single Assignment), register allocation, and intermediate program representations;

• the design of high-level program analysis, optimizations, and tools, mainly related to high-level synthesis, some leading to the development of the Zettice start-up.

For more details on the past years of Compsys II, see the previous annual reports from 2008 to 2012. Compsys II was positively evaluated in Spring 2012 by Inria. The evaluation committee members were Walid Najjar (University of California Riverside), Paolo Faraboschi (HP Labs), Scott Mahlke (University of Michigan), Pedro Diniz (University of Southern California), Peter Marwedel (TU Dortmund), and Pierre Paulin (STMicroelectronics, Canada), the last three assigned specifically to Compsys.

For Compsys III, the changes in the permanent members (departure of Fabrice Rastello and arrival of Laure Gonnord (while she was only external collaborator of Compsys until Sep. 2013) reduces the forces on back-end code optimizations, and in particular dynamic compilation, but increases the forces on program analysis. In this context, Compsys III will continue to develop fundamental concepts or techniques whose applicability should go beyond a particular architectural or language trend, as well as stand-alone tools (either as proofs of concepts or to be used as basic blocks in larger tools/compilers developed by others) and our own experimental prototypes. One of the main objectives of Compsys III is to try to push the polyhedral model beyond its present limits both in terms of analysis techniques (possibly integrating approximation and runtime support) and of applicability (e.g., analysis of parallel or streaming languages, program verification, compilation towards accelerators such as GPU or multicores).