Reliable Generation of High-Performance Matrix Algebra

Author:

Nelson Thomas1,Belter Geoffrey1,Siek Jeremy G.2,Jessup Elizabeth1,Norris Boyana3

Affiliation:

1. University of Colorado

2. Indiana University

3. University of Oregon

Abstract

Scientific programmers often turn to vendor-tuned Basic Linear Algebra Subprograms (BLAS) to obtain portable high performance. However, many numerical algorithms require several BLAS calls in sequence, and those successive calls do not achieve optimal performance. The entire sequence needs to be optimized in concert. Instead of vendor-tuned BLAS, a programmer could start with source code in Fortran or C (e.g., based on the Netlib BLAS) and use a state-of-the-art optimizing compiler. However, our experiments show that optimizing compilers often attain only one-quarter of the performance of hand-optimized code. In this article, we present a domain-specific compiler for matrix kernels, the Build to Order BLAS (BTO), that reliably achieves high performance using a scalable search algorithm for choosing the best combination of loop fusion, array contraction, and multithreading for data parallelism. The BTO compiler generates code that is between 16% slower and 39% faster than hand-optimized code.

Funder

NSF

Publisher

Association for Computing Machinery (ACM)

Subject

Applied Mathematics,Software

Reference43 articles.

1. S. Amarasinghe D. Campbell W. Carlson etal 2009. Exascale software study: Software challenges in extreme scale systems. Tech. Rep. DARPA IPTO Air Force Research Labs. S. Amarasinghe D. Campbell W. Carlson et al. 2009. Exascale software study: Software challenges in extreme scale systems. Tech. Rep. DARPA IPTO Air Force Research Labs.

2. Achieving high sustained performance in an unstructured mesh CFD application

3. Can search algorithms save large-scale automatic performance tuning? Procedia;Balaprakash P.;Comput. Sci. CS,2011

4. Synthesis of High-Performance Parallel Programs for a Class of ab Initio Quantum Chemistry Models

Cited by 8 articles. 订阅此论文施引文献 订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献

1. A Tensor Algebra Compiler for Sparse Differentiation;2024 IEEE/ACM International Symposium on Code Generation and Optimization (CGO);2024-03-02

2. Optimizing Tensor Programs on Flexible Storage;Proceedings of the ACM on Management of Data;2023-05-26

3. EGGS: Sparsity‐Specific Code Generation;Computer Graphics Forum;2020-08

4. SPIRAL: Extreme Performance Portability;Proceedings of the IEEE;2018-11

5. Format abstraction for sparse tensor algebra compilers;Proceedings of the ACM on Programming Languages;2018-10-24

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3