Scholarlink

<p>We present an optimised distributed-memory algorithm for the unpreconditioned conjugate gradient (CG) method. We experiment with using the CG algorithm with the compressed sparse row (CSR), compressed sparse row&#xa0;5 (CSR5), compressed sparse row&#xa0;2 (CSR2) and symmetric compressed sparse row (SCSR) sparse matrix formats. For the CG algorithm, we present a CSR5-based and an SCSR-based distributed-memory sparse matrix-vector multiplication (SpMV) algorithm. We developed five distributed-memory CG implementations using different combinations of the mentioned sparse matrix formats and SpMV algorithms. We benchmarked our implementations on a distributed-memory computer cluster with Intel x86-64 CPUs. For the maximum benchmarked number of compute nodes, the median performance of our optimised distributed-memory CG algorithm with CSR-based SpMV was 58&#xa0;%&#xa0;greater than of the state-of-the-art PETSc library, and the median performance of our optimised distributed-memory CG algorithm with SCSR-based SpMV was 105&#xa0;%&#xa0;greater than of the PETSc library.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Accelerating the Conjugate Gradient Method on Distributed-Memory Computers

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月16日

查看全文
  • Claudio Kozický,
  • Ivan Šimeček,
  • Boris Rúra

摘要

We present an optimised distributed-memory algorithm for the unpreconditioned conjugate gradient (CG) method. We experiment with using the CG algorithm with the compressed sparse row (CSR), compressed sparse row 5 (CSR5), compressed sparse row 2 (CSR2) and symmetric compressed sparse row (SCSR) sparse matrix formats. For the CG algorithm, we present a CSR5-based and an SCSR-based distributed-memory sparse matrix-vector multiplication (SpMV) algorithm. We developed five distributed-memory CG implementations using different combinations of the mentioned sparse matrix formats and SpMV algorithms. We benchmarked our implementations on a distributed-memory computer cluster with Intel x86-64 CPUs. For the maximum benchmarked number of compute nodes, the median performance of our optimised distributed-memory CG algorithm with CSR-based SpMV was 58 % greater than of the state-of-the-art PETSc library, and the median performance of our optimised distributed-memory CG algorithm with SCSR-based SpMV was 105 % greater than of the PETSc library.