(A) two-way memory interleaved SIMD machine for image processing

Abstract

In this thesis, a two-way memory interleaved SIMD machine is proposed. This model can solve the incompatibility between processing elements (PEs) and the size of the problem which is larger than the number of PEs. It can also achieve memory interleaving effect for the purpose of utilization of full utilization of the PE``s. The algorithms for such problems that require relatively simple (e.g. 3 by 3 neighborhood processing) and relatively complicated (e.g. FFT, histogram calculation) patterns of data transfer are described. And the data placement rules for this model are also described.