Sophie, an FDTD code on the way to multicore, getting rid of the memory bandwidth bottleneck better using cache

Abstract

FDTD codes, such as Sophie developed at CEA/DAM, no longer take advantage of the processor's increased computing power, especially recently with the raising multicore technology. This is rooted in the fact that low order numerical schemes need an important memory bandwidth to bring and store the computed fields. The aim of this article is to present a programming method at the software's architecture level that improves the memory access pattern in order to reuse data in cache instead of constantly accessing RAM memory. We will exhibit a more than two computing time improvement in practical applications. The target audience of this article is made of computing scientists and of electrical engineers that develop simulation codes with no specific knowledge in computer science or electronics.