Fast sparse matrix-vector multiplication by exploiting variable block structure

  • Authors:

  • Richard W. Vuduc;Hyun-Jin Moon

  • Affiliations:

  • Lawrence Livermore National Laboratory;University of California, Los Angeles

  • Venue:
  • HPCC'05 Proceedings of the First international conference on High Performance Computing and Communications
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

We improve the performance of sparse matrix-vector multiplication(SpMV) on modern cache-based superscalar machines when the matrix structure consists of multiple, irregularly aligned rectangular blocks. Matrices from finite element modeling applications often have this structure. We split the matrix, A, into a sum, A1 + A2 + ... + As, where each term is stored in a new data structure we refer to as unaligned block compressed sparse row (UBCSR) format. A classical approach which stores A in a BCSR can also reduce execution time, but the improvements may be limited because BCSR imposes an alignment of the matrix non-zeros that leads to extra work from filled-in zeros. Combining splitting with UBCSR reduces this extra work while retaining the generally lower memory bandwidth requirements and register-level tiling opportunities of BCSR. We show speedups can be as high as 2.1× over no blocking, and as high as 1.8× over BCSR as used in prior work on a set of application matrices. Even when performance does not improve significantly, split UBCSR usually reduces matrix storage.