FFT FOR THE APE PARALLEL COMPUTER

We present a parallel FFT algorithm for SIMD systems following the ''T ranspose Algorithm'' approach. The method is based on the assignment o f the data field onto a one-dimensional ring of systolic cells. The sy stolic array can be universally mapped onto any parallel system. In pa rticular for systems with next-neighbor connectivity our method has th e potential to improve the efficiency of matrix transposition by use o f hyper-systolic communication. We have realized a scalable parallel F FT on the APE100/Quadrics massively parallel computer, where our imple mentation is part of a two-dimensional hydrodynamics code for turbulen ce studies.