Domain specific compiler for coordinated signal processing in 5G testbed

Abstract

In the past years, the Internet of Things (IoT) is changing our working mode and life. In order to address the demand of massive data, low latency and low power consumption in IoT applications, the wireless industry is moving to its fifth generation (5G). Due to high computational requirements in the 5G, DSP(Digital Signal Processor) will play an important role in the 5G and IoT. Swift DSP is a novel high-performance DSP, with the support of SIMD (Signal Instruction Multiple Data) and VLIW (Very Long Instruction Word) to achieve high parallelism. In order to implement application on Swift based 5G testbed, compile tools are needed. Because of the unique architecture feature of Swift, there is no existing compile tools appropriate for Swift. Thus, we implement a set of optimized compile tools for Swift to exploit parallelism while developing applications. It uses C as programming language which is general and easy for users to develop and implement applications. Finally, we evaluate the performance of our VLIW packetizer, SIMD vectorization. For VLIW, it improves performance by 56% because almost every program will have some independent instructions. SIMD achieves 35% performance improvement since data in 5G is always high-dimensional. The combination of them gains 55% to 65% on average.