OptiSDR — A domain specific language to improve developer productivity for software defined radio

Abstract

Domain specific languages (DSLs) offer the potential for using a small, specialized and highly expressive programming language to concisely represent solutions for a particular domain of problems. The DSL approach is in stark contrast to that of a general programming language that attempts to provide a means to represent solutions for a broad range of problems. DSLs may incorporate constructs and syntactical features that not only relate to the problem context, but also to the execution platform. DSLs thus provide a potential for reducing development time and thereby cutting business costs. At present this approach is likely more appropriate to prototyping and rapid experimental work; but as technologies to implement DSLs improve they may become more beneficial and cost effective for developing marketable products. In this paper we present a DSL, called OptiSDR, aimed to improve developer productivity for development of software defined radio (SDR) applications deployed on heterogeneous computing architectures (HCAs). OptiSDR is designed for HCAs that adhere to certain combinations of processing devices, namely FPGAs, graphics processing units, and multi-core general purpose processors. Our aim to improve developer productivity in multiple ways was achieved, with development cycles for parallel DSP routines for SDR reduced to few hours as opposed to days for CUDA-C development cycles. OptiSDR uses automatic code generation for structured parallel SDR patterns, and performs automatic profiling to generate faster execution kernels with speed-ups of more than 50× compared to other sequential implementations while reducing manual effort in porting and tuning code. Therefore this paper reports on a study confirming the potential to speed up when using OptiSDR for prototyping SDR applications.