From structured to object oriented programming in parallel algorithms for 3D image reconstruction

TL;DR: This paper shows how the abstractions employed to model a problem, may play an important role from the performance and scalability perspectives.

Abstract: Emerging computer architectures intend to overcome limitations imposed by the power consumption, thermal issues and physics. Multicore processors and computing platforms where GPUs and CPUs are combined, are turning into attractive computing platforms for developers. These architectures are characterized for breaking the trend in computing. They are single computing entities built on an aggregate of smaller and pseudo independent processing silicon dies. New trends in parallel computing are shifting towards the multicore arena. A new factor arises in such scenario, the concurrence. But concurrence is not parallelism. Parallel applications (and sequential ones, of course) that wish to take advantage of this new environment need to take this into consideration, or being completely rewritten in such a way that parallelism can be expressed by means of concurrence. In this changing scenario, abstractions may help to keep performance. This paper shows how the abstractions employed to model a problem, may play an important role from the performance and scalability perspectives.