The DeepGreen project aims at deploying Artificial Intelligence solutions (AI) for specific Hardware (HW) targets provided by French companies including Small and Medium Sized companies (SMEs). In particular, a primary objective is to design and develop a versatile software development platform able to generate code for the specific HW constraints of each targeted solutions.
Additional requirements on the provided software developement platform are: efficient and sound support for embedded targets, debugability and fine control, sustainability and frugality, open and robust source code. The ambition is to put the French companies’ components and IPs in a competivite position.
The CORSE Team contributes to the design and implementation of the compiler infrastructure, intermediate representations and exchange formats, the performance model for dense and sparse computations, the code generators and auto tuning framework.
The OTPaaS project targets massive documents digitization by offering a suitable cloud platform compatible with the Gaia-X project and easy to use by companies including SMEs. The consortium brings together national technology providers and users from major groups and SMEs/ETIs, with strong support from major French research institutes. The platform OTPaaS will be validated by 6 demonstrators and followed by ambitious industrialization programs.
The CORSE Team contributes to energy efficiency and auto tuning of the generated code for the targeted cloud and edge devices.
The main ambition of the ES3CAP project is to develop a software stack both for multi- and many-cores platforms that target the market of critical systems. One of the main objective is to provide software solutions that will allow existing critical systems to scale, providing high performance, real-time constraints, reliablility, and security. Targeted markets include aviation and drone industry, defense, and autonomous driving. The projects plans to prototype its solutions on the Kalray manycore architecture.
Back-end industrial partners include Renault-Nissan and EasyMile for the automotive sector, in addition to Safran E&D and MBDA for the aviation and defense sectors. Other partners include: ANSYS / Esterel Technologies (tools), Krono-Safe (software reliability and real-time constraints), Prove & Run (security of operating system), Kalray (software/hardware platform MPPA), Prophesee (bio-inspired sensors and AI), IROC Tech (hardware reliability and security). Academic partners are Inria and CEA.
Nano2017 PSAIC (2015-2017)
The PSAIC project concerns the automation of program optimization through the combination of several tools and techniques such as: compiler optimization, profiling, trace analysis, and iterative optimization. For any given application, the objective is to devise through a fully automated process a compiler profile optimized for performance and code size. For this purpose, we are developing instrumentation techniques that can be focused and specialized to a specific part of the application aimed to be monitored. One of the goals is to gain order of magnitudes in the profiling overhead and analysis cost. To this end, we propose to develop: 1. a new adaptative monitoring technique; 2. a hybrid behavior analysis; 3. extensions of hybrid compilation techniques (with region cloning). Developments will be integrated in ATOS auto-tuning tool, GCC and LLVM compiler.