Parallelize Accelerated Triangle Counting Using Bit-Wise on GPU

Li Lin, Dian Ouyang, Zhipeng He, Chengqian Li

Abstract: Triangle counting is a graph algorithm that calculates the number of triangles in a graph, the number of triangles is a key metric for a large number of graph algorithms. Traditional triangle counting algorithms are divided into vertex-iterator and edge-iterator when traversing the graph. As the scale of graph data grows, the use of CPU with other architectural platforms for triangle counting has become mainstream. Our accelerating method proposes an algorithm for triangle counting on a single machine GPU, and performs a two-dimensional partition algorithm for large-scale graph data, in order to ensure that large graph data can be correctly loaded into GPU memory and the independence of each partition to obtain the right result. According to the high concurrency of GPU, a bit-wise operation intersection algorithm is proposed. We experiment with our algorithm to verify that our method effectively speeds up triangle counting algorithm on a single-machine GPU.