Yet Another Survey on SIMD Instructions

Abstract

Data level parallelism approach enables a potential speed-up on applications that works with (almost) independent data sets. Recently, parallel processing has been growing and has being commonly used in different applications, such as image/video processing, visualization, scientific computation, etcetera. The latest processor’s micro-architectures come equipped with special units that process an operation over a set of data, on a SIMD fashion. In this document, we research main features offered by different instruction sets that perform parallel operations. We focus on various instruction sets designed for either desktop/server computers such as: SSEx, AVX and FMA, and also for mobile and embedded computers with ARM-based processors, which supports the NEON instruction set.

Citation

@unpublished{faz_yassimd2013,
  author = {Armando Faz-Hernández},
  title = {Yet Another Survey on SIMD Instructions},
  pages = {1-8},
  year = {2013},
  month = {jun}
}