Complete Stream Fusion for Software-Defined Radio

Abstract

Strymonas is a code-generation--based library (embedded DSL) for fast, bulk, single-thread in-memory stream processing -- with the declarative description of stream pipelines and yet achieving the speed and memory efficiency of hand-written state machines. It guarantees complete stream fusion in all cases. So far, strymonas has been used on small examples and micro-benchmarks. In this work, we evaluate strymonas on a large, real-life application of Software-Defined Radio -- FM Radio reception, -- contrasting and benchmarking it against the synchronous dataflow system StreamIt, and the state-of-the art: GNU Radio. Strymonas, despite being declarative, single-thread single-core with no explicit support for SIMD, no built-in windowing or convolution, turns out to offer portable high performance, well enough for real-time FM Radio reception. It is on par with (or, on Raspberry Pi Zero, outstripping) GNU Radio, while providing static guarantees of complete fusion and type safety.