Abstract: We consider automatic parallelization of a computational kernel executed according to the PRedictable Execution Model (PREM), where each thread is divided into execution and memory phases.