A symbolic cost model for asynchronous parallel programs with structured dependences
Abstract
We propose to associate a symbolic cost to a asynchronous parallel program. In contrast to classical approaches, we take into account all the dynamic asynchronism related to the different initial states that generate dynamic independences. We introduce an asynchronous intermediate model called SCP. It only imposes static independences allowing a sequential lecture of programs. To compute the symbolic cost, we condition the existence of a dependence by annotating it with a predicate on program variables. To evaluate the cost in the initial environment, predicates are transformed by a backward method based on the model Weakest Liberal Preconditions Calculus. The resulting symbolic cost is parameterized by the initial environment and integrates dynamic dependences.