A Fast, Vectorizable Algorithm for Producing Single-Precision Sine-Cosine Pairs

TL;DR

This paper introduces a fast, vectorizable algorithm for computing sine and cosine pairs in single-precision, optimized for architectures like PowerPC AltiVec and easily adaptable to others such as Intel SSE.

Contribution

The paper proposes a novel, branch-free algorithm for sine-cosine computation that enhances performance through vectorization and is portable across different processor architectures.

Findings

High-speed sine-cosine pair computation without branches

Efficient implementation on PowerPC AltiVec processors

Easy adaptation to architectures like Intel SSE

Abstract

This paper presents an algorithm for computing Sine-Cosine pairs to modest accuracy, but in a manner which contains no conditional tests or branching, making it highly amenable to vectorization. An exemplary implementation for PowerPC AltiVec processors is included, but the algorithm should be easily portable to other achitectures, such as Intel SSE.