FLOPs as a Discriminant for Dense Linear Algebra Algorithms

被引:0
|
作者
Lopez, Francisco [1 ]
Karlsson, Lars [1 ]
Bientinesi, Paolo [1 ]
机构
[1] Umea Univ, Umea, Sweden
来源
51ST INTERNATIONAL CONFERENCE ON PARALLEL PROCESSING, ICPP 2022 | 2022年
关键词
linear algebra; algorithm selection; scientific computing; SET;
D O I
10.1145/3545008.3545072
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Expressions that involve matrices and vectors, known as linear algebra expressions, are commonly evaluated through a sequence of invocations to highly optimised kernels provided in libraries such as BLAS and LAPACK. A sequence of kernels represents an algorithm, and in general, because of associativity, algebraic identities, and multiple kernels, one expression can be evaluated via many different algorithms. These algorithms are all mathematically equivalent (i.e., in exact arithmetic, they all compute the same result), but often differ noticeably in terms of execution time. When faced with a decision, high-level languages, libraries, and tools such as Julia, Armadillo, and Linnea choose by selecting the algorithm that minimises the FLOP count. In this paper, we test the validity of the FLOP count as a discriminant for dense linear algebra algorithms, analysing "anomalies": problem instances for which the fastest algorithm does not perform the least number of FLOPs. To do so, we focused on relatively simple expressions and analysed when and why anomalies occurred. We found that anomalies exist and tend to cluster into large contiguous regions. For one expression anomalies were rare, whereas for the other they were abundant. We conclude that FLOPs is not a sufficiently dependable discriminant even when building algorithms with highly optimised kernels. Plus, most of the anomalies remained as such even after filtering out the inter-kernel cache effects. We conjecture that combining FLOP counts with kernel performance models will significantly improve our ability to choose optimal algorithms.
引用
收藏
页数:10
相关论文
共 50 条
  • [21] Profiling high performance dense linear algebra algorithms on multicore architectures for power and energy efficiency
    Ltaief, Hatem
    Luszczek, Piotr
    Dongarra, Jack
    COMPUTER SCIENCE-RESEARCH AND DEVELOPMENT, 2012, 27 (04): : 277 - 287
  • [22] Energy Footprint of Advanced Dense Numerical Linear Algebra using Tile Algorithms on Multicore Architectures
    Dongarra, Jack
    Ltaief, Hatem
    Luszczek, Piotr
    Weaver, Vincent M.
    SECOND INTERNATIONAL CONFERENCE ON CLOUD AND GREEN COMPUTING / SECOND INTERNATIONAL CONFERENCE ON SOCIAL COMPUTING AND ITS APPLICATIONS (CGC/SCA 2012), 2012, : 274 - 281
  • [23] Energy-efficient execution of dense linear algebra algorithms on multi-core processors
    Alonso, Pedro
    Dolz, Manuel F.
    Mayo, Rafael
    Quintana-Orti, Enrique S.
    CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS, 2013, 16 (03): : 497 - 509
  • [24] Scalable Parallel Approach for Dense Linear Algebra
    Abouelfarag, Ahmed A.
    Nouh, Nada Magdy
    ElShenawy, Marwa
    2016 INTERNATIONAL CONFERENCE ON HIGH PERFORMANCE COMPUTING & SIMULATION (HPCS 2016), 2016, : 1003 - 1008
  • [25] Mechanizing the Expert Dense Linear Algebra Developer
    Marker, Bryan
    van de Geijn, Robert
    Batory, Don
    Terrel, Andy
    Poulson, Jack
    ACM SIGPLAN NOTICES, 2012, 47 (08) : 289 - 290
  • [26] Accelerating GPU Kernels for Dense Linear Algebra
    Nath, Rajib
    Tomov, Stanimire
    Dongarra, Jack
    HIGH PERFORMANCE COMPUTING FOR COMPUTATIONAL SCIENCE - VECPAR 2010, 2011, 6449 : 83 - 92
  • [27] BLOCK-CYCLIC DENSE LINEAR ALGEBRA
    LICHTENSTEIN, W
    JOHNSSON, SL
    SIAM JOURNAL ON SCIENTIFIC COMPUTING, 1993, 14 (06): : 1259 - 1288
  • [28] Benchmarking GPUs to Tune Dense Linear Algebra
    Volkov, Vasily
    Demmel, James W.
    INTERNATIONAL CONFERENCE FOR HIGH PERFORMANCE COMPUTING, NETWORKING, STORAGE AND ANALYSIS, 2008, : 499 - 509
  • [29] Looking back at dense linear algebra software
    Luszczek, Piotr
    Kurzak, Jakub
    Dongarra, Jack
    JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING, 2014, 74 (07) : 2548 - 2560
  • [30] Randomized algorithms in numerical linear algebra
    Kannan, Ravindran
    Vempala, Santosh
    ACTA NUMERICA, 2017, 26 : 95 - 135