Applied Mathematics & Information Sciences
The use of many-core architectures improves the performance of most of the data-intensive applications. One of the challenging tasks for modern many-core architectures is to handle the irregular memory access effectively. Unlike regular memory access applications, an increase in the size of the problem in an irregular memory access application leads to a reduction in overall performance. A mapping between on-chip and off-chip memory through the heterogeneous communication channel also poses significant challenges. In this paper, a k-exchange algorithm with ant colony optimization is proposed to improve the performance of irregular memory access applications such as Multi-dimensional Knapsack Problem (MKP) and Traveling Salesman Problem (TSP) on the Graphics Processing Units (GPU). A different set of instances of OR library and TSPLIB are considered for experiments. The obtained results show an improvement in terms of optimal solution and speedup for the MKP and TSP instances.
Digital Object Identifier (DOI)
P. S., Tamizharasan and N., Ramasubramanian
"Enhanced Data Parallelism for Irregular Memory Access Optimization on GPU,"
Applied Mathematics & Information Sciences: Vol. 13:
4, Article 11.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol13/iss4/11