ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2010.05975
35
7

On the Parallel I/O Optimality of Linear Algebra Kernels: Near-Optimal LU Factorization

12 October 2020
Grzegorz Kwa'sniewski
Tal Ben-Nun
A. Ziogas
Timo Schneider
Maciej Besta
Torsten Hoefler
ArXivPDFHTML
Abstract

Dense linear algebra kernels, such as linear solvers or tensor contractions, are fundamental components of many scientific computing applications. In this work, we present a novel method of deriving parallel I/O lower bounds for this broad family of programs. Based on the X-partitioning abstraction, our method explicitly captures inter-statement dependencies. Applying our analysis to LU factorization, we derive COnfLUX, an LU algorithm with the parallel I/O cost of N3/(PM)N^3 / (P \sqrt{M})N3/(PM​) communicated elements per processor -- only 1/3×1/3\times1/3× over our established lower bound. We evaluate COnfLUX on various problem sizes, demonstrating empirical results that match our theoretical analysis, communicating asymptotically less than Cray ScaLAPACK or SLATE, and outperforming the asymptotically-optimal CANDMC library. Running on 111,024024024 nodes of Piz Daint, COnfLUX communicates 1.6×\times× less than the second-best implementation and is expected to communicate 2.1×\times× less on a full-scale run on Summit.

View on arXiv
Comments on this paper