Memory-oriented optimization techniques in General Purpose GPU programming

dc.contributor.author	Pasternak, Nazar
dc.date.accessioned	2024-02-14T09:34:09Z
dc.date.available	2024-02-14T09:34:09Z
dc.date.issued	2022
dc.identifier.citation	Pasternak, Nazar. Memory-oriented optimization techniques in General Purpose GPU programming / Pasternak, Nazar; Supervisor: Oleg Farenyuk; Ukrainian Catholic University, Department of Computer Sciences. – Lviv: 2022. – 41 p.
dc.identifier.uri	https://er.ucu.edu.ua/handle/1/4416
dc.language.iso	en	uk
dc.title	Memory-oriented optimization techniques in General Purpose GPU programming	uk
dc.type	Preprint	uk
dc.status	Публікується вперше	uk
dc.description.abstracten	Effective utilization of the GPU parallel execution potential for GPGPU solutions requires an extensive understanding of the internal GPU memory model. GPU’s memory latency hiding techniques differ from those of the CPU, due to substantial design differences. In this thesis, we review the specifics of the GPU memory hierarchy, identify potential memory bottlenecks of GPGPU programs and address them using the CUDA programming model. We provide examples of such solutions along with corresponding performance measurements. As a demonstration of the proposed optimizations, we provide the implementation of the Parallel Failureless Aho-Corasick algorithm for pattern-matching and measure the performance speedup each optimization resulted in. Optimizations discussed in this paper result in almost 2x performance speedup of highly memory-dependant PFAC algorithm.	uk