Toward a model of computation for time-constrained applications on manycores
Abstract
As computing systems are transitioning from multicores to manycores with the increasing number of computing resources available on modern chips, we can notice a lack of a universal programming model for these new platforms and the challenges they convey. Ideally speaking, a program should be written only once, and making it run on a given target would be the role of the compilation tools. But before addressing this problem, we need a good Model of Computation (MoC) as a base for both programming and compilation tools. In this paper we propose to share our insights on the properties such a MoC should possess. It would take the CycloStatic DataFlow (CSDF) MoC for its good properties, and extend it to overcome its limitations while retaining its good properties.