A modified parallel approach to Single Source Shortest Path Problem for massively dense graphs using CUDA

Today´s Graphics Processing Units (GPUs) possess enormous computation power with highly parallel and multithreaded architecture, and the most attractive feature being their low costs. NVIDIA´s CUDA provides an interface to the developers to use the GPUs for General Purpose Parallel Computing. In this paper, we present a modified algorithm of Single Source Shortest Path Problem on GPU using CUDA. First, we modify the standard BELLMAN-FORD algorithm to remove its drawbacks and make it suitable for parallel implementation, and then implement it using CUDA. For dense graphs, our Algorithm gains a speedup of 10×-12× over the previously proposed parallel algorithm. Our Algorithm also accept graphs with negative weighted edges and can detect any reachable Negative Weighted Cycle, which widens its scope to accept generalized problems.