Optimizing the Performance of Streaming Numerical Kernels on the IBM Blue Gene/P PowerPC 450 Processor

摘要

Several emerging petascale architectures use energy-efficient processors withvectorized computational units and in-order thread processing. On thesearchitectures the sustained performance of streaming numerical kernels,ubiquitous in the solution of partial differential equations, represents achallenge despite the regularity of memory access. Sophisticated optimizationtechniques are required to fully utilize the Central Processing Unit (CPU). We propose a new method for constructing streaming numerical kernels using ahigh-level assembly synthesis and optimization framework. We describe animplementation of this method in Python targeting the IBM Blue Gene/Psupercomputer's PowerPC 450 core. This paper details the high-level design,construction, simulation, verification, and analysis of these kernels utilizinga subset of the CPU's instruction set. We demonstrate the effectiveness of our approach by implementing severalthree-dimensional stencil kernels over a variety of cached memory scenarios andanalyzing the mechanically scheduled variants, including a 27-point stencilachieving a 1.7x speedup over the best previously published results.