Comparison of parallel computing approaches of a finite-difference implementation of the acoustic diffusion equation model

摘要

The diffusion equation model has been intensively researched as a room-acoustics simulation algorithm during last years. A 3-D finite-difference implementation of this model was proposed to evaluate the propagation over time of sound field within rooms. Despite the computational saving of this model to calculate the room energy impulse response, elapsed times are still long when high spatial resolutions and/or simulations in several frequency bands are needed. In this work, several data-parallel approaches of this finite-difference solution on Graphics Processing Units are proposed using a compute unified device architecture programming model. A comparison of their performance running on different models of Nvidia GPUs is carried out. In general, 2D vertical block approach running in a Tesla K20C shows the best speed-up of more than 15 times versus CPU version.