Support for irregular computations in massively parallel PIM arrays, using an object-based execution model

Abstract

The emergence of semiconductor fabrication technology allowing a tight coupling betw een high-density DRAM and CMOS logic on the same chip has led to the important new class of Processor-In-Memory (PIM) architectures. Furthermore, large arrays of PIMs can be arranged into massively parallel architectures. In this paper, we outline the salient features of PIM architectures and discuss macroservers, an object-based model for such machines. Subsequently, we specifically address the support for irregular problems provided by PIM arrays. The discussion concludes with a case study illustrating an approach to the solution of a sparse matrix vector multiplication. ?© 2000 Springer-Verlag Berlin Heidelberg.