Section: Software
ANR MORE platform
Participants : Jonathan Ponroy, Olivier Zendra.
With three partners (LIP6 in Paris, IRIT in Toulouse, and INRIA-LORIA), the ANR MORE (Multicriteria Optimization for Real-time Embedded systems) project aims at developing trade-off strategies that transform the code of a critical embedded application so that it meets the system constraints in terms of worst-case execution time, code size and energy consumption. In this project, we at INRIA Nancy Grand Est focus on memory optimizations for energy under real-time constraints. With our partners, we are developing in this project an iterative optimization process that will help in driving the selection of the transformations to apply according to measures carried on the system (through a simulator provided by the IRIT partner).
In the past years, we were laying the foundations for this platform. Thanks to the recruitment of Jonathan Ponroy this year, we have been able to make very significant progress in this area and worked on building the software bases for the experimental platform. More precisely, we have been developing measure tools for the energy part and for memory characterization (initially loosely based on SPECO and other developments, but not anymore) and have been working on their integration with and into the libraries ands simulator (Otawa) provided by our IRIT partner, and with the code compression extension provided by our LIP6 partner. We completed the memory cache behavior of Otawa, taking into account new elements, such as its write policy. Scratchpad memory and a part of DRAM memory were also added to Otawa to more accurately simulate energy consumption.
We are now able to simulate architectures with several memory banks, various memory placement strategies for data, and obtain the relevant energy consumption. To test and visualize theses results, we automated the simulation task, stressing various architectural environments and memory placements. We also automated the generation of graphs to visualize this large mass of data directly in OpenOffice. This helps us for the exploitation of the experimental results for the scientific work itself.
Finally, once these bases were solidly established, we started working on the iterative, multicriteria process part per se, with a strong interaction with LIP6. The architectural work for this part of the platform is still ongoing, since for example we want to be able to easily change the exploration algorithm.
