Efficient reduction over threads

• Main Author: Falkman, Patrik
• Format: Others
• Language: English
• Published: KTH, Teoretisk fysik 2011
• Subjects: Computer and Information Sciences; Data- och informationsvetenskap
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-49818

The increasing number of cores in both desktops and servers leads to a demand for efficient parallel algorithms. This project focuses on the fundamental collective operation reduce, which merges several arrays into one by applying a binary operation element wise. Several reduce algorithms are evaluated in terms of performance and scalability and a novel algorithm is introduced that takes advantage of shared memory and exploits load imbalance. To do so, the concept of dynamic pair generation is introduced which implies constructing a binary reduce tree dynamically based on the order of thread arrival, where pairs are formed in a lock-free manner. We conclude that the dynamic algorithm, given enough spread in the arriving times, can outperform the reference algorithms for some or all array sizes.