PolyTop++: an efficient alternative for serial and parallel topology optimization on CPUs & GPUs

This paper presents the PolyTop++, an efficient and modular framework for parallel structural topology optimization using polygonal meshes. It consists of a C++ and CUDA (a parallel computing model for GPUs) alternative implementations of the PolyTop code by Talischi et al. (Struct Multidiscip Optim...

Full description

Saved in:
Bibliographic Details
Published in:Structural and multidisciplinary optimization 2015-11, Vol.52 (5), p.845-859
Main Authors: Duarte, Leonardo S., Celes, Waldemar, Pereira, Anderson, M. Menezes, Ivan F., Paulino, Glaucio H.
Format: Article
Language:English
Subjects:
Algorithms
C (programming language)
Computational Mathematics and Numerical Analysis
Engineering
Engineering Design
Finite element method
Iterative methods
Memory management
Modular structures
Research Paper
Solvers
Theoretical and Applied Mechanics
Topology optimization
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper presents the PolyTop++, an efficient and modular framework for parallel structural topology optimization using polygonal meshes. It consists of a C++ and CUDA (a parallel computing model for GPUs) alternative implementations of the PolyTop code by Talischi et al. (Struct Multidiscip Optim 45(3):329–357 2012b ). PolyTop++ was designed to support both CPU and GPU parallel solutions. The software takes advantage of the C++ programming language and the CUDA model to design algorithms with efficient memory management, capable of solving large-scale problems, and uses its object-oriented flexibility in order to provide a modular scheme. We describe our implementation of different solvers for the finite element analysis, including both direct and iterative solvers, and an iterative ‘matrix-free’ solver; these were all implemented in serial and parallel modes, including a GPU version. Finally, we present numerical results for problems with about 40 million degrees of freedom both in 2D and 3D.
ISSN:1615-147X
1615-1488
DOI:10.1007/s00158-015-1252-x