Fine-grained Parallel ILU Preconditioners with Fill-ins for Multi-core CPUs and GPUs
SRG New Frontiers in High Performance Computing, Engineering Mathematics and Computing Lab (EMCL), Karlsruhe Institute of Technology (KIT), Germany
International Conference On Preconditioning Techniques For Scientific And Industrial Applications, 2011
@article{lukarski2011fine,
title={Fine-grained Parallel ILU Preconditioners with Fill-ins for Multi-core CPUs and GPUs},
author={Lukarski, D. and Heuveline, V. and Weiss, J.P. and others},
year={2011}
}
Numerical simulation and its huge computational demands require a close coupling between efficient mathematical methods and their hardware-aware implementation on emerging and highly parallel computing platforms. The paradigm shift towards manycore parallelism not only offers a high potential of computing capabilities but also comes up with urgent challenges in designing scalable, portable, and flexible software solutions. The latter point is closely related to adaptation, variation, and restructuring of algorithms and numerical schemes in order to be compliant with coarse- and fine-grained parallelism, hierarchical memory subsystems, heterogeneous platforms, and any kind of communication bottleneck. Numerical codes should not only be efficient and robust, but also future-proof with respect to the current dynamic landscape of hardware platforms and parallel programming environments. The goal of our work is to treat problems with great impact on science and humanities, e.g. in the domain of meteorology, medical engineering and energy research, can be modeled and solved. With efficient preconditioners the bounds of solvability can be pushed further beyond.
October 4, 2011 by hgpu
