Section: Software

Introduction

This section lists and briefly describes the software developments conducted within Compsys. Most are tools that we extend and maintain over the years. They now concern two activities only: a) the development of tools linked to polyhedra and loop/array transformations, b) the development of algorithms within the back-end compiler of stm icroelectronics.

The previous annual reports contain descriptions of Pip, a tool for parametric integer linear programming, of the Polylib tool, a C library of polyhedral operations, and of MMAlpha a circuit synthesis tool for systolic arrays. These tools, developed by members or past members of Compsys, are not maintained anymore in Compsys, but are extended by other groups. They have been important tools in the development of the “polytope model”, which is now widely accepted: it is used by Inria projects-teams Cairn and Alchemy, PIPS at École des Mines de Paris, Suif from Stanford University, Compaan at Berkeley and Leiden, PiCo from the hp  l abs (continued as PicoExpress by Synfora), the dtse methodology at Imec, Sadayappan's group at Ohio State University, Rajopadhye's group at Colorado State's University, etc. These groups are research projects, but the increased involvement of industry (Hewlett Packard, Philips, Synfora, Reservoir Labs) is a favorable factor. Polyhedra are also used in test and certification projects (Verimag, Lande, Vertecs). More recently, several compiler groups have shown their interest in polyhedral methods: the gcc group and Reservoir Labs in the USA, which develops a compiler fully-based on the polytope model and on the techniques we introduced for loop and array transformations. Now that these techniques are well-established and disseminated by other groups (in particular Alchemy), we prefer to focus on the development of new techniques and tools, which are described here.

The other activity concerns the developments within the compiler of stm icroelectronics. These are not stand-alone tools, which could be used externally, but algorithms and data structures implemented inside the LAO back-end compiler, year after year, with the help of stm icroelectronics colleagues. As these are also important developments, it is worth mentioning them in this section. They are also completed by important efforts for integration and evaluation within the complete stm icroelectronics toolchain. They concern exact methods (ILP-based), algorithms for aggressive optimizations, techniques for just-in-time compilation, and for improving the design of the compiler.