In the third quarter of 2017 Intel released CPUs with the AVX-512 SIMD instruction set. These instructions include the processing of vectors of 16 integers in parallel rather than one at a time - and all in a single instruction. We are interested in how these instructions might be used to improve the performance of a search engine. In this talk we'll outline how our experimental search engine works, and areas that we believe can be improved by using SIMD instructions (our on-going research). In particular we'll be discussing SIMD-based compression of monotonically increasing integer sequences and why it's important.
Last modified: Monday, 28-May-2018 15:30:28 NZST
This page is maintained by the seminar list administrator.