A distributed memory algorithm for 3-D FFTs (Q1919366)

This paper proposes an algorithm for the parallel in-place 3-D fast Fourier transform (FFT) on distributed memory architectures. In this algorithm, calculation is done by partitioning the 3-D input tensor into subtensors (one subtensor per nodes). The local FFTs are computed by the divide-and-conquer method, and the results are combined along the data partitioning strategy for minimum communication between nodes. The complexity of this algorithm is discussed in detail.