Study of BDRM Asynchronous Parallel Computing Model Based on Multiple CUDA Streams

In order to improve the computing speed of ocean acoustic field using the Beam-Displacement Ray-Mode (BDRM) theory, a BDRM parallel computing model based on Compute Unified Device Architecture (CUDA) is designed by virtue of the powerful parallel computing ability of GPU and the character of BDRM theory. The emphasis is how to implement parallel computing of eigen value and eigen function in CUDA programming model. The results of simulation experiment show that the CPU elapsed time increases fast but the GPU elapsed time increases slow with the frequency of the sound source reaching higher. The speedup in blue-water is bigger than that in shallow-water under the same frequency of the sound source. The speedups are 7.84× and 33.36× respectively in shallow-water and blue-water when the frequency of the sound source is 1000Hz. The BDRM parallel computing model based on CUDA has higher computing efficiency than the BDRM serial computing model based on CPU under large scale operations. It could achieve the requirement of fast forecast of ocean acoustic field and engineering application.