Optimization by runtime specialization for sparse matrix-vector multiplication

Xu, Danqing

Permalink

https://hdl.handle.net/2142/72903 Copy

Description

Title

Optimization by runtime specialization for sparse matrix-vector multiplication

Author(s)

Xu, Danqing

Issue Date

2015-01-21

Director of Research (if dissertation) or Advisor (if thesis)

Garzaran, Maria J.

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Computer Science
• Parallel Programming
• Program Optimization
• Code generation
• Runtime Specialization
• Programming Languages

Abstract

Runtime specialization optimizes programs based on partial information available only at run time. It is applicable when some input data is used repeatedly while other input data varies. This technique has the potential of generating highly efficient codes. In this thesis we explore the potential for obtaining speed-ups for sparse matrix-dense vector multipli- cation using runtime specialization, in the case where a single matrix is to be multiplied by many vectors. We experiment with five methods involving run-time specialization with parallelization, comparing them to methods that do not (including Intel’s MKL library). For this work, our focus is the evaluation of the parallel speed-ups that can be obtained with runtime specialization without considering the overheads of the code generation. Our experiments run on four different machines with 88 matrices from the Matrix Market and Florida collections, among others. In 348 of those 352 cases, the specialized code runs faster than any version without specialization. In the worst case, the specialized code is 7 percent slower than the Intel’s MKL library. If we only use specialization, the average speedup with respect to Intel’s MKL library ranges from 1.416x to 1.470x, depending on the machine. We have also found that the best method depends on the matrix and machine; no method is best for all matrices and machines.

Graduation Semester

2014-12

Permalink

http://hdl.handle.net/2142/72903

Copyright and License Information

Copyright 2014 Danqing Xu

Owning Collections