In this paper, we present a partitioning methodology targeting a dynamically reconfigurable architecture. We first identify this class of architectures and point out the lack of underlying tools and compiler support exploiting, with these architectures, the potential Task Level Parallelism (TLP).

The applications in today?s and future embedded systems are more and more control dominated making necessary the use of a specification that handles jointly treatment and control. Our methodology starts from a system level specification in Safe State Machines (SSM: the graphical formalism of ESTEREL) integrating task level granularity treatments (as C function calls) in a control flow environment. After simulation and formal proof, we explicitly partition all the different ?configurations? of the SSM i.e. the different combinations of control and treatment that the system has to perform at each operational tick. We also develop a technique that enables to contain the explosion of the number of these ?configurations? and establish the efficiency of our method through its application to a video supervision application and to the JPEG 2000 standard.