In this presentation , we will discuss how a compiler can automatically detect the parallelism present in an application and how it can be expressed in a convenient format based on a process network formalism. This formalism makes it easy to partition code and map it on heterogeneous multicore platforms.