A fast fourier transformation algorithm for single-chip cloud computers using RCCE

Multimedia applications, spectrum analyses and data compression algorithms employ Fourier transformations as one of their main components to transform series from the time to the spectral domain and vice versa. Effective fast Fourier transformation (FFT) algorithms imply performance enhancements in related applications. Although architecturespecific programs achieve better performance, many FFT implementations were designed to be hardware independent and unaware of the underlining architecture. In this paper we introduce a novel FFT algorithm based on the RCCE native message passing library for the single-chip cloud computer (SCC). We parallelized the recursive (divide-and-conquer) radix- 2 FFT such that the inputs for all processing units are independent. Private memories are used to avoid cache coherence issues, and the algorithm was designed to minimize the message passing overhead. Preliminary experimental results were conducted using the RCCE emulator on an Intel Xeon 2 CPU quad-core computer. The emulator results showed promising scalability and speed-ups over the sequential implementation. Based on hardware availability, we plan to run the experiments on real SCC hardware for the final version of this paper.