Fast, good, and repeatable: Summations, vectorization, and reproducibility. (September 2020)

• Record Type: Journal Article
• Authors: Neuman, Brett; Dubois, Andy; Monroe, Laura; Robey, Robert W
• Other Names: Mascagni Michael guest-editor.
• Is Part Of: International journal of high performance computing applications. Volume 34:Number 5(2020)
• Page Start: 519
• Page End: 531
• Publication Date: 2020-09
• Subjects: Reproducibility -- vectorization -- self-compensated summation -- enhanced precision -- reproducible sums; High performance computing -- Periodicals; Supercomputers -- Periodicals; 004.1105
• DOI: 10.1177/1094342020938425
• Languages: English
• ISSNs: 1094-3420

Abstract

Enhanced-precision global sums are key to reproducibility in exascale applications. We examine two classic summation algorithms and show that vectorized versions are fast, good and reproducible at exascale. Both 256-bit and 512-bit implementations speed up the operation by almost a factor of four over the serial version. They thus demonstrate improved performance on global summations while retaining the numerical reproducibility of these methods.