PROGRAM SYSTEMS: THEORY AND APPLICATIONS

12+

Online Scientific Journal published by the Ailamazyan Program Systems Institute of the Russian Academy of Sciences

Papers are accepted in the form of a PDF file

To view the PDF files, you will need Adobe Acrobat Reader

• Содержание выпуска •
• Mathematical Modelling •
• Mathematical Foundations of Programming •
• Supercomputing Software and Hardware •
• Software and Hardware for Distributed Systems and Supercomputers •
• Artificial Intelligence, Intelligence Systems, Neural Networks •

Supercomputing Software and Hardware

Responsible for the Section: Sergei Abramov, Dr. Phys.-Math.Sci., corresponding member of RAS

On the left: assigned number of the paper, submission date, the number of A5 pages contained in the paper,
and the reference to the full-text PDF .

Article # 13_2020 27 с. PDF	submitted on 29th Apr 2020 displayed on website on 20th Aug 2020 Konstantin Isupov, Vladimir Knyazkov Multiple-precision matrix-vector multiplication on graphics processing units We are considering a parallel implementation of matrix-vector multiplication (GEMV, Level 2 of the BLAS) for graphics processing units (GPUs) using multiple-precision arithmetic based on the residue number system. In our GEMV implementation, element-wise operations with multiple-precision vectors and matrices consist of several parts, each of which is calculated by a separate CUDA kernel. This feature eliminates branch divergence when performing sequential parts of multiple-precision operations and allows the full utilization of the GPU’s resources. An efficient data structure for storing arrays with multiple-precision entries provides a coalesced access pattern to the GPU global memory. We have performed a rounding error analysis and derived error bounds for the proposed GEMV implementation. Experimental results show the high efficiency of the proposed solution compared to existing high-precision packages deployed on GPU. (in Russian). Key words: Top500, supercomputer, interconnect, hybrid architectures.
article citation	http://psta.psiras.ru/read/psta2020_3_33-59.pdf
DOI	https://doi.org/10.25209/2079-3316-2020-11-3-33-59
Article # 14_2020 24 с. PDF	submitted on 29th Apr 2020 displayed on website on 20th Aug 2020 Konstantin Isupov, Vladimir Knyazkov Multiple-precision matrix-vector multiplication on graphics processing units We are considering a parallel implementation of matrix-vector multiplication (GEMV, Level 2 of the BLAS) for graphics processing units (GPUs) using multiple-precision arithmetic based on the residue number system. In our GEMV implementation, element-wise operations with multiple-precision vectors and matrices consist of several parts, each of which is calculated by a separate CUDA kernel. This feature eliminates branch divergence when performing sequential parts of multiple-precision operations and allows the full utilization of the GPU’s resources. An efficient data structure for storing arrays with multiple-precision entries provides a coalesced access pattern to the GPU global memory. We have performed a rounding error analysis and derived error bounds for the proposed GEMV implementation. Experimental results show the high efficiency of the proposed solution compared to existing high-precision packages deployed on GPU. Key words: Top500, supercomputer, interconnect, hybrid architectures.
article citation	http://psta.psiras.ru/read/psta2020_3_61-84.pdf
DOI	https://doi.org/10.25209/2079-3316-2020-11-3-61-84

Adress: Ailamazyan Program Systems Institute of the Russian Academy of Sciences, PSTA Online Journal, 4 a Peter the First Street,
Veskovo village, Pereslavl area, Yaroslavl region, 152021 Russia
Phone: +7-4852-695-228. E-mail: info@psta.psiras.ru. Website: http://psta.psiras.ru