Pseudo-vector machine for embedded applications

摘要: Recently much attention has been focused on designing low-cost, low-power and high performance processors for mid-to-low end embedded mobile applications. In this work, we will present a new processing paradigm, called the Pseudo-Vector Machine, such particular, focus how to improve program loop executions by adding small amount of hardware single issued machine. The characteristics critical loops in these applications vary greatly, from highly vectorizable, difficult costly vectorize, impossible vectorize. For vector arithmetic that can be represented vectorizable loops, traditional paradigm typically provides largest gains. hard vectorize DSP's style which focuses optimizations, is more suitable. pseudo-vector machine proposed dissertation perform two types single, integrated datapath. Depending nature sometimes behaves like processor; it DSP processor. The compiler, machine, decides best suited given loop. By providing ability execute both paradigms achieve improvements are higher than either individual paradigm. average speedup, case, over set applications, 1.731.