OCR extensions - local identifiers, labeled GUIDs, file IO, and data block partitioning

TL;DR

This paper proposes several extensions to the OCR specification, including local identifiers, labeled GUIDs, file IO, and data block partitioning, to improve efficiency, safety, and flexibility in parallel computing environments.

Contribution

It introduces new OCR extensions such as local validity identifiers, labeled GUIDs, file IO interface, and data block partitioning for enhanced performance and safety.

Findings

Local identifiers enable more optimization opportunities.

Labeled GUIDs facilitate safe concurrent object creation.

Data block partitioning improves access control.

Abstract

We present several proposals for extending the Open Community Runtime (OCR) specification. The extension are identifiers with local validity, which use the concept of futures to provide OCR implementations more optimization opportunities, labeled GUIDs with creator functions, which are based on the local identifiers and allow the developer to create arrays of OCR objects that are safe from race conditions in case of concurrent creation of objects, a simple file IO interface, which builds on top of the existing data block concepts, and finally data block partitioning, which allows better control and flexibility in situations where multiple tasks want to access disjoint parts of a data block.