Implementation of data manipulating functions on the staran associative processor

Abstract

Parallel processing data manipulating functions have been widely discussed in the literature. These functions are used to move data from one processing element to another in a parallel manner. A large subset of these functions (shift, flip, shuffle, spread, replicate, close-up) have been implemented as macros on the STARAN S-1000 Associative Processor. An end off shift was found to require 3W to 8W instruction cycles (W is the field width in bits). Other functions have been implemented requiring log₂N (N=string length) end off shift operations. Execution times are minimum when string lengths are a power of two. Data manipulation execution times are typically greater than the time required for an array add instruction but much less than required for an array multiply. Aspects of the STARAN architecture as related to the data manipulating functions are discussed.