Accelerating Industrial Applications: The Development of Basic GPU Kernels for the New Block AMG Algorithms for Solving SLE with Explicitly Calculated Sparse Basis

详细信息    查看全文

• 作者：Ilya Afanasyev^a ; ¹ ; ^{Afanasiev_ilya@icloud.com" class="auth_mail" title="E-mail the corresponding author} ; Yury Potapov^a ; ² ; ^{potap.yurich@gmail.com" class="auth_mail" title="E-mail the corresponding author} ; Sergey Sobolev^a ; ³ ; Sergey Kharchenko^b ; ⁴
• 关键词：CUDA ; performance optimizations ; linear algebra
• 刊名：Procedia Computer Science
• 出版年：2015
• 出版时间：2015
• 年：2015
• 卷：66
• 期：Complete
• 页码：642-650
• 全文大小：415 K

文摘

Nowadays, GPU computations are playing significant role in supercomputing technologies. This work is a part of a project dealing with solving problems of modeling hydro- and aerodynamics where linear algebra operations are frequently used and occupy most of execution time. In despite of the fact that GPUs are traditionally used for solving high sized problems, in our project we need to solve many tasks of low sizes. Because of this, modern library's solutions such as cuBLAS (1) and cuSPARSE (2) are not suitable enough for that, so we have a task of implementation more efficient functions for concrete linear algebra operations taking into account its specialties.