CISL Brown Bag Seminar: "Stability Analysis of Inline ZFP Compression for Floating-Point Data in Iterative Methods"

09/05/2018 - 12:00pm to 1:00pm
ML 680 Tower B (Penthouse)

"Stability Analysis of Inline ZFP Compression for Floating-Point Data in Iterative Methods"

Alyson Fox
Lawrence Livermore National Lab

ZFP, a state-of-the-art lossy compression algorithm, can easily be used inline during numerical simulations due to the inherent locality of the algorithm. ZFP decomposes a solution state into 4d blocks, which are subsequently compressed (decompressed) independently. In a numerical simulation, the solution state already contains traditional errors, e.g., floating-point round-off, truncation error, and discretization error. The information that is lost during ZFP compression may represent the traditional errors, however, any additional error caused by ZFP could contaminate the current iterate. It is important to understand if the error from ZFP compression overwhelms the other sources of error. The goal of this work is to analyze the stability of using ZFP in fixed precision inline in time-evolving iterative methods. We introduce the infinite bit vector space and define corresponding operators for each step of ZFP, to establish an error bound for both a static and iterative implementation of ZFP in fixed precision. To conclude, we present results from several numerical experiments to observe the validity of our established error bounds.

Bring your lunch!

Wednesday, September 5, 2018
12:00pm - 1:00pm
ML, Room 680 Tower B (Penthouse)