by David Martinez, Vasco Santos, Martin Mc Donnell, Ken Reynolds, and Peter Carlston, Intel Corporation
The suitability of Intel® multi-core processors for embedded digital signal processing (DSP) applications is now being reevaluated. Major advances in power-efficient transistor technology, optimized multi-core processor microarchitectures and the evolution of Intel® Streaming SIMD Extensions (Intel® SSE) for vector processing have combined to produce favorable GFLOPS/watt and GFLOPS/size ratios. In addition, other factors such as code portability across the entire range of Intel® processors and a large set of Intel and third-party software development tools and performance libraries often mean that software development and support costs can be substantially reduced.
This article explores the main differences between traditional digital signal processors and modern Intel general purpose processor architectures and gives some orientation on how DSP engineers can most effectively take advantage of the resources available in Intel processors. We then show how these techniques were used to implement and benchmark performance of medical ultrasound, wireless infrastructure, and advanced radar processing algorithms on a variety of current Intel processors.