Section: Overall Objectives
The overall objective of the CASH team is to take advantage of the characteristics of the specific hardware (generic hardware, hardware accelerators, or reconfigurable chips) to compile energy efficient software and hardware. More precisely, we plan to work on:
Definition of dataflow representations of parallel programs that can capture the parallelism at all levels: fine-grain vs. coarse-grain, data & task parallelism, programming language, and intermediate representation (Section 3.1).
Scalable and expressive static program analyses. CASH will work on improving the scalability of analyses to allow a global analysis of large-scale programs, and on the expressiveness of analysis to find better program invariants. Analysis will be performed both on the representation defined above and on general programs (Section 3.2).
Transformations from and to the dataflow representation, combining traditional tools dedicated to dataflow and specific methods like the polyhedral model (Section 3.3).
A high-level synthesis (HLS) tool, built on the above item (instantiated with the particularities of FPGAs) and a code generation tool (Section 3.4). This HLS tool will focus on early stages of compilation and rely on an external tool for the back-end.
A parallel and scalable simulation of hardware systems, which, combined with the preceding activity, will result in an end-to-end workflow for circuit design (Section 3.5).
To ensure the coherency and the correctness of our approach these different tasks will rely on a precise definition of the manipulated languages and their semantics. The formalization of the different representations of the programs and of the analyses will allow us to show that these different tasks will be performed with the same understanding of the program semantics.
Note that these directions are strongly tied together. We use 5 research axis for the sake of the presentation, but their complementarity enables each member of the team to share common research goals while having their own research directions. Most of our results will contribute to several directions.