FUZZY NUMBER SUBTRACTION CONVOLUTION ON THE CM-2
Nikolay Petkov · International Journal of Modern Physics C · 1993
The running time of data-parallel algorithms can be reduced by avoiding global reduction operations. Using fuzzy number subtraction convolution as an example, this paper illustrates how data-parallel programs can be restructured to eliminate global reduction operations. This is done by using techniques from systolic array design. The algorithms are programmed in CM Fortran and have been run on a Connection Machine CM-2. The running time results show that advantages can be achieved in a considerable domain of the problem size space. At the same time the results show that for obtaining shortest running times different algorithms have to be used in different domains.