| 摘要 |
A method of performing an integral transform operation (such as a Fast Fourier Transform), wherein the underlying algorithm is partitioned to provide an efficient sequence of data operations. Preferably the address calculations are performed separately from the data calculations, and the algorithm is partitioned so that the microcode sequence for all but the last few data calculations is constant. Thus, the bandwidth at the interface to the numeric processor is conserved, and control storage in the numeric processor is also efficiently conserved. Moreover, the preferred partition for performing Fast Fourier Transform manipulates data in reasonably large subsets (e.g. 8 floating-point words at a time). This turns out to use less data bandwidth than would be required using smaller data subsets. |
