Optimization Trade-offs in the Synthesis of Software for Embedded DSP

A broad variety of implementation considerations become important when implementing software for embedded signal processing applications. These include complex trade-offs such as those involving code size, data buffering requirements, performance, and power consumption. This presentation motivates the use of high-level, dataflow-based programming models for automated synthesis of DSP software; presents an overview of optimization issues and trade-offs that arise in uniprocessor software synthesis from dataflow specifications; describes useful techniques that have been developed to address some of these trade-offs; and motivates important problems in this area for further investigation.