Description

The MNEMEE project aims at developing source-to-source optimization methodologies and tools to improve the design of MPSoC embedded systems, realizing ambitious future applications. The proposed optimizations make possible the mapping of very demanding applications and increase the cost efficiency of the final mapping. The optimization methodologies and tools are divided in two parts: i) compiler-independent part and ii) memory-hierarchy aware (but processor architecture independent) part. In particular, the first part is addressed by adaptive source-to-source optimizations in dynamically allocated data (see WP2) and optimizations for statically allocated data through scenarios (see WP3). The second part is addressed by source-to-source optimizations for the placement of all data in the distributed/shared data memory hierarchy of a MPSoC platform (see WP4). The need to have source-to-source optimization methodologies for statically and dynam ically allocated data comes from the fact that the statically allocated data are allocated at compile time and the dynamically allocated data are allocated and deallocated at run-time. Therefore, WP2 and WP3 have to address completely different challenges and thus, completely different optimization flows and strategies are required from them. The optimizations in both parts aim at the improvement of specific cost factors; more specifically, decrease of data access related execution time, decrease of energy consumption, reduction of memory footprint and increase of effectively available memory bandwidth.
Figure 1 shows the relation between the first five Workpackages. The first Workpackage delivers data analysis, memory hierarchy requirements, scenario identification and specifications for the embedded software applications. The second, third and fourth Workpackages provide the source-to-source optimizations. The fifth Workpackage provides the demonstrators for the MNEMEE optimization approach.

Complete Workpackage List

WP1: Data analysis, specifications and scenario identification for embedded software applications (M1-M12)
Workpackage Leader: Dimitrios Kritharidis, INTRACOM

WP2: Source-to-source optimizations of dynamically allocated data mapping on MPSoC platforms (M3-M24)
Workpackage Leader: Dimitrios Soudris, ICCS

WP3: Source-to-source optimizations of statically allocated data mapping on MPSoC platforms (M6-M30)
Workpackage Leader: Maryse Wouters, IMEC

WP4: Memory hierarchy-aware optimization techniques (M3-M27)
Workpackage Leader: Peter Marwedel, ICD

WP5: Demonstrators (M18-M36)
Workpackage Leader: Francois Capman, THALES Communications

WP6: Dissemination and exploitation (M1-M36)
Workpackage Leader: Maryse Wouters, IMEC

WP7: Project Management (M1-M36)
Workpackage Leader: Stylianos Mamagkakis, Technical Project Advisor, IMEC