Parallel Three-Dimensional Fast Fourier TransformsDeveloped by Dmitry Pekurovsky, SDSC Scientific Computing Applications Group
Taking Fourier Transforms of data in three dimensions is a fairly common operation in many scientific codes. This is usually a compute intensive operation that can take a significant part of the total run time of some codes. This SAC/SCC project consists of open source software addressing three-dimensional fourier transforms in parallel (dubbed P3DFFT). The goal is to remove bottlenecks to scaling in preparation for petascale computing. As such P3DFFT implements two-dimensional (or pencil) domain decomposition, in contrast to many existing approaches that implement one-dimensional decomposition. This removes a barrier to scaling inherent in the one-dimensional approach, namely that the number of processors could not exceed the grid size of the physical problem in any one dimension.
Download the latest version description and P3DFFT library >