Toward a Portable Programming Environment for Distributed High Performance Accelerators

摘要

Accelerators with little power consumption per computation performance are beginning to widely spread for High Performance Computing use, instead of general-purpose CPUs with much power consumption. They are GPUs, processors of Cell architecture, and FPGA accelerators. While these processors have much higher computation performance than general-purpose CPUs, they need specific programming environment respectively when using them as distributed memory accelerators. We discuss a portable programming environment which can be used in common with distributed memory accelerators in this paper.