Research on GPU-Based Computation Method for Line-of-Sight Queries

The line of sight (LOS) queries often consume a significant fraction of system resources in military simulation. The high time complexity of LOS computation not only limit the amount of entities in the simulation, but also hamper the CPU from doing more urgent and important tasks. To overcome this problem, we utilize graphic process units (GPU) to accelerate the LOS computation at two levels, single-query level and batch-query level. First, we decouple the dependency of data to parallelize the whole process of LOS computation, so that the potential of GPU can be exploited at single-query level. Second, a combine-and-partition algorithm is proposed to aggregate multiple single LOS queries into a GPU-based computation, so that the count of parallel threads can be maximized and the impact of communication latency can be minimized. It uses a combine function to assemble scattered single query into a batch query, and uses a partition function to get computational data or dispatch results. An early version of our prototype demonstrates at least 3× speedup at single-query level, and we expect to achieve a speedup eyond 200× at batch-query level based on the LOS culling methods in references 1 and 2.