The use of overlapping subgrids to accelerate the FDTD on GPU devices

Abstract

The method Finite Difference Time Domain (FDTD) is widely used in electromagnetic simulations to solve problems of microwave tomography, radar and telecommunications. Since this method is a data intensive and computation intensive problem, there are a lot of initiatives to improve the scalability and the performance of the FDTD. Despite the progress, performance in FDTD simulation is still a challenge, especially in situations with experiments with large environments, complex objects and/or electromagnetic waves with a wide range of frequencies. The use of GPU to accelerate the FDTD, which has a good cost-benefit, is especially in focus, offering a speedup of hundreds of times if compared to the traditional CPU computation. The memory manager is the main issue to improve the performance of FDTD over GPU. This work presents a solution which uses overlapping subgrids in order to increase the efficiency in the memory access. The performed tests show a speedup of 1.7 times of the proposed solution when compared to the current FDTD over GPU.

Efficient utilization of memory hierarchy to enable the computation on bigger domains for stencil computation in CPU-GPU based systems

“…In order to reduce the communication cost between CPU and GPU side, temporal blocking method, where each sub-domain is computed for multiple time steps at once without frequent communication, has been introduced [5], [6], [7] as shown in Figure 4. We name TBS as the number of time steps computed at once on the GPU side.…”

“…So, it has to copy current sub-domain back to CPU side and copy the next sub-domain to GPU side to continue which cause frequent communication between CPU and GPU. There is temporal blocking method [5], [6], [7], [8] which can solve this frequent communication problem. It has been utilized to improve cache locality [9], [10], [11] and to reduce the number of inter process communication [12].…”