Search
 
loader loading subjects...
cover
Design Towards Modern High Performance Numerical LA Library Enabling Heterogeneity and Flexible Data Formats
Toshiyuki Imamura, Daichi Mukunoki, Yusuke Hirota, Susumu Yamada, Masahiko Machida
97 - 106
10.3233/978-1-61499-843-3-97
Advances in Parallel Computing
Volume 32: Parallel Computing is Everywhere
Abstract

This work introduces a design proposal towards modernization of high performance numerical library enabling various parallel execution, such as offloading, many-core concurrent execution, heterogeneous hybrid execution. A prototype implementation of Eigen-G2 employs a parallel task model supported by OpenMP, and exclusive control of offloading a GPU device. Also, multiple data formats are available by taking advantage of a metaprogramming support of C++ language. Eigen-G2 exhibits good parallel performance scalability when we use both multicore CPUs and a GPU at the same time.

$35.00 / €27.50 / £22.00
Add PDF to cart