Sorting on GPUs for large scale datasets: A thorough comparison
Information Processing and Management: an International Journal
Hi-index | 0.00 |
This paper presents a new multicasting network constructed with a bit-level cost complexity of O(N log/sup 2/N) and a bit-level time complexity of O(log/sup 2/ N) using comparators with bit-level I(1) time and cost complexities. The requested addresses for connection are sorted in a pipeline fashion (worm-hole routed) bit-serially most-significant-bit first through the network Bitonic sorter.The sorted addresses and the source nodes' addresses are merged through a Bitonic merger. The comparators and their links in both the Bitonic sorter and merger are bi-directional so two-way paths are formed.