Performing global transforms on an SIMD machine
This paper addresses the problem of performing global transforms of images on an SIMD machine proposed for image processing . In particular, parallel computation of two-dimensional fast Fourier transform (2-D FFT) is considered. Two of the machine's permissible memory accesses are exploited, which allow the N processing elements of the machine to access two halves of two rows or columns of an N×N image in the memory system in one memory access. It is shown that in two memory accesses, each processing element is able to obtain two pixels of one of the two rows or columns which are 2n−1 pixels apart. This access capability in conjunction with the machine's permutation network lead to an efficient algorithm for parallel computation of 2-D FFT through row-column decomposition.
KeywordsSIMD machine permutation network global transforms 2-D FFT
Unable to display preview. Download preview PDF.
- 1.K.E. Batcher, “The flip network in STARAN,” Proc. Int'l Conf. Parallel Processing, pp. 65–71, Aug. 1976.Google Scholar
- 2.P.E. Danielsson and S. Levialdi, “Computer architectures for pictorial information systems,” Computer, vol. 14, pp. 53–67, Nov. 1981.Google Scholar
- 4.D.-L. Lee, A multiple-processor architecture for image processing, Ph.D. Thesis, Department of Computing Science, University of Alberta, 1987.Google Scholar
- 6.A.P. Reeves, “Parallel computer architectures for image processing,” Computer vision, Graphics, and Image Processing, vol. 25, pp. 68–88, Feb. 1984.Google Scholar
- 7.A. Rosenfeld and A.C. Kak, Digital picture processing, Academic Press, 1982.Google Scholar
- 8.H.S. Stone, “Parallel processing with perfect shuffle,” IEEE Trans. Comput., vol. C-20, pp. 153–161, Feb. 1971.Google Scholar