Fast sparse matrix-vector multiplication by exploiting variable block structure

PDF Version Also Available for Download.

Description

We improve the performance of sparse matrix-vector multiply (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 kind of structure. Our technique splits the matrix, A, into a sum, A{sub 1} + A{sub 2} + ... + A{sub s}, where each term is stored in a new data structure, unaligned block compressed sparse row (UBCSR) format . The classical alternative approach of storing A in a block compressed sparse row (BCSR) format yields limited performance gains because it imposes a particular alignment of ... continued below

Physical Description

PDF-file: 32 pages; size: 1 Mbytes

Creation Information

Vuduc, R W & Moon, H July 7, 2005.

Context

This report is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided by UNT Libraries Government Documents Department to Digital Library, a digital repository hosted by the UNT Libraries. It has been viewed 11 times , with 5 in the last month . More information about this report can be viewed below.

Who

People and organizations associated with either the creation of this report or its content.

Publisher

Provided By

UNT Libraries Government Documents Department

Serving as both a federal and a state depository library, the UNT Libraries Government Documents Department maintains millions of items in a variety of formats. The department is a member of the FDLP Content Partnerships Program and an Affiliated Archive of the National Archives.

Contact Us

What

Descriptive information to help identify this report. Follow the links below to find similar items on the Digital Library.

Description

We improve the performance of sparse matrix-vector multiply (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 kind of structure. Our technique splits the matrix, A, into a sum, A{sub 1} + A{sub 2} + ... + A{sub s}, where each term is stored in a new data structure, unaligned block compressed sparse row (UBCSR) format . The classical alternative approach of storing A in a block compressed sparse row (BCSR) format yields limited performance gains because it imposes a particular alignment of the matrix non-zero structure, leading to extra work from explicitly padded zeros. Combining splitting and UBCSR reduces this extra work while retaining the generally lower memory bandwidth requirements and register-level tiling opportunities of BCSR. Using application test matrices, we show empirically that speedups can be as high as 2.1x over not blocking at all, and as high as 1.8x over the standard BCSR implementation used in prior work. When performance does not improve, split UBCSR can still significantly reduce matrix storage. Through extensive experiments, we further show that the empirically optimal number of splittings s and the block size for each matrix term A{sub i} will in practice depend on the matrix and hardware platform. Our data lay a foundation for future development of fully automated methods for tuning these parameters.

Physical Description

PDF-file: 32 pages; size: 1 Mbytes

Language

Item Type

Identifier

Unique identifying numbers for this report in the Digital Library or other systems.

  • Report No.: UCRL-TR-213454
  • Grant Number: W-7405-ENG-48
  • DOI: 10.2172/891708 | External Link
  • Office of Scientific & Technical Information Report Number: 891708
  • Archival Resource Key: ark:/67531/metadc875745

Collections

This report is part of the following collection of related materials.

Office of Scientific & Technical Information Technical Reports

Reports, articles and other documents harvested from the Office of Scientific and Technical Information.

Office of Scientific and Technical Information (OSTI) is the Department of Energy (DOE) office that collects, preserves, and disseminates DOE-sponsored research and development (R&D) results that are the outcomes of R&D projects or other funded activities at DOE labs and facilities nationwide and grantees at universities and other institutions.

What responsibilities do I have when using this report?

When

Dates and time periods associated with this report.

Creation Date

  • July 7, 2005

Added to The UNT Digital Library

  • Sept. 21, 2016, 2:29 a.m.

Description Last Updated

  • Nov. 28, 2016, 2:55 p.m.

Usage Statistics

When was this report last used?

Yesterday: 0
Past 30 days: 5
Total Uses: 11

Interact With This Report

Here are some suggestions for what to do next.

Start Reading

PDF Version Also Available for Download.

International Image Interoperability Framework

IIF Logo

We support the IIIF Presentation API

Vuduc, R W & Moon, H. Fast sparse matrix-vector multiplication by exploiting variable block structure, report, July 7, 2005; Livermore, California. (digital.library.unt.edu/ark:/67531/metadc875745/: accessed October 15, 2018), University of North Texas Libraries, Digital Library, digital.library.unt.edu; crediting UNT Libraries Government Documents Department.