Trading performance for memory in sparse direct solvers using low-rank compression

Loris Marchal, Thibault Marette, Grégoire Pichon, Frédéric Vivien

Abstract: Highlights•We propose a memory-aware strategy to enhance low-rank compression.•We show that the formalization of the problem is a variant of the Knapsack problem.•We design models to estimate the information needed by the approximation algorithm.•We implement a dynamic version of the memory-aware strategy in the PaStiX solver.