Reasoning about Inherent Parallelism in Object-Oriented Programs using Ownership Types

The principal premise of this paper is that as a field, we do not currently have a suitable conceptual framework for reasoning about inherent parallelism. We have techniques called dependence analysis for determining the safety of parallelization, but these techniques do not provide a basis for abstraction and so do not scale well to entire applications that are large, complex and developed from components. This paper seeks to extend both the principles and practice of parallel programming by proposing a new abstraction for reasoning about inherent parallelism and uses that abstraction to develop a set of sufficient conditions for data parallel loops