Object-Oriented Data Parallel Programming in C++

A lot of applications executed on distributed memory parallel computers can be classi-ed as data parallel applications, in which parallel operations on data elements are performed in a loosely synchronous manner. However implementation of such applications is still diicult. This paper presents an object-oriented approach for an easy and eecient data parallel programming in C++. More exactly, an object model for distributed data is deened for modularity, polymorphism, and object sharing in programs, and data distribution and message passing details are encapsulated to reduce programming complexity and make reuse of code easier. Experiments show that this approach is ee-cient for a large class of data parallel applications.