Section: Highlights of the Year

Highlights of the Year

Scientific Results and Dissemination

Despite the approaching end of Compsys, we continued the objectives we fixed for Compsys III, i.e., pushing static compilation beyond its present limits, both in terms of techniques and applications. Our most important efforts in 2016 were to extend static analysis from sequential codes to parallel specifications and languages, to develop polynomial techniques, and to increase inter-disciplinary collaborations and dissemination towards HPC users and their applications. The most important results in 2016 are the following:

• Publications Well recognized in the polyhedral community, we got three papers at IMPACT'16, the central event of this community, one paper at the main compiler conference (CC'16), and a last one in the field of FPGA, which remains an important target for polyhedral optimizations. See Sections 7.1 to 7.7 for more details.

• Interdisciplinary spring school With colleagues from HPC numerical simulation, we organized a very successful inter-disciplinary event in May 2016, to bridge the gap between polyhedral compilation and HPC users. See details in Section 10.1.

• Move towards HPC users In addition to the spring school we organized, we increased our activity towards HPC users and their applications through the supervision of the internship of J. Versaci (quantum physics), the reviewing of T. Gasc's PhD thesis (fluid dynamics), and the regular contacts with the LMGC lab (mechanics).

• PhD theses The end of Compsys coincided also with the end of two PhD theses, the PhD thesis of Guillaume Iooss [16] and the PhD thesis of Alexandre Isoard [17], see Section 10.2.2.

• Final evaluation The team was evaluated in March 2016, this was also its final evaluation.

Final Evaluation and End of Compsys

Compsys has been created in 2002 as an Inria team, then in 2004 as an Inria project-team, and evaluated by Inria first in 2007, then in 2012. It was evaluated again in March 2016, which was its final evaluation because an Inria project-team is limited to 12 years. The construction of a new project was planned in early 2015, following the shift in the research directions that started in the second half of Compsys III. A few tentative research directions were:

• Shift the application domain from embedded systems to high performance computing (HPC) but at small scale (desktop HPC: FPGA, GPU, multicores). In fact, the two ecosystems are nowadays slowly converging.

• A stronger attention to real HPC users and real HPC applications may lead to better programming models (“putting the programmer in the loop”).

• Design new models of programs. The polynomial model is but an example.

• Explore the synergy between parallel programming and program verification and certification; in particular, import approximation methods from one field to the other. Abstract interpretation is a case in point.

However, while its field of expertise, compilation for parallel and heterogeneous systems, is still of crucial importance, the unexpected departure in Sep. 2015 of two of its staff members made this future impossible. We nevertheless continued in 2016, in particular to present our activities in this last evaluation, until the three last members had to split in three different cities (Lyon, Paris, Rennes). We report here some of the comments made by the external reviewers that, we think, summarize well some aspects of our efforts, successes, and difficulties during 15 years:

• Compsys established and matured the polyhedral optimization approach, which is the state of the art for locality and parallelism optimization in optimizing compilers. The project has had world-wide impact.

• We strongly recommend that the members of the team are accommodated in Camus, Cairn, Parkas, or another complementary Inria team, irrespective of the geographical location. Otherwise, Inria will lose one of its peaks of research excellence in Computer Science.

• This team is a prime example where Inria requirements on teams are damaging science and collaboration.

• This team has produced many impactful results and is considered as the Polyhedral center of excellence. It is globally recognized for its research in both front-end (polyhedral optimizations) and back-end (graph optimizations) compiler optimization techniques integrating elegant foundational theory with real implementation on various architectures (multi-core, FPGAs, DSP, GPU etc.).

• In back‐end optimizations, the team had developed the state-of‐the‐art SSA and decoupled register allocation techniques that are important to achieving peak performance.

• They have internationally visible and impactful research in compilers, technology transfer to companies through collaborations and through start‐ups. They raised the global awareness of polyhedral analysis through creation of workshops, summer schools etc., essentially reviving interest in the topic about a decade ago, and finally educating next‐generation of researchers in this area, who are now contributing to both academic and industrial research landscape in France and beyond.

• The start‐up company (XtremLogic on HLS) is an excellent concrete evidence of technology transfer from the team. [...] In the future, a more careful analysis of the trade-off between technology transfer and academic research is necessary for small project teams so that a promising research direction does not get jeopardized in Inria.

• The Compsys team has truly achieved research excellence in compilation techniques. Unfortunately, the future of the team remains uncertain due to administrative policies. Inria should enable the team to continue with their research strengths in polyhedral analysis and graph‐theory based SSA-type optimizations.