Vectorizing Irregular Programs

Dnr:

SNIC 2017/3-70

Type:

SNAC Small

Principal Investigator:

Kim-Anh Tran

Affiliation:

Uppsala universitet

Start Date:

2017-08-07

End Date:

2018-09-01

Primary Classification:

10206: Datorteknik

Webpage:

Allocation

Abstract

Vectorizing irregular programs still imposes a big challenge for compiler writers. In order to expand the capabilities of auto-vectorization, we want to explore the new features of the AVX-512 extension, and find out how they can help to speculatively execute vector instructions. Introducing speculation may increase the scope of application for vectorization, and thus improve program performance.