The Community for Technology Leaders
Green Image
<p>A discussion is presented of the use of dynamic storage schemes to improve parallelmemory performance during three important classes of data accesses: vector accesses inwhich multiple strides are used to access a single vector, block accesses, andconstant-geometry FFT accesses. The schemes investigated are based on linear addresstransformations, also known as XOR schemes. It has been shown that this class ofschemes can be implemented more efficiently in hardware and has more flexibility thanschemes based on row rotations or other techniques. Several analytical results areshown. These include: quantitative analysis of buffering effects in pipelined memorysystems; design rules for storage schemes that provide conflict-free access usingmultiple strides, blocks, and FFT access patterns; and an analysis of the effects ofmemory bank cycle time on storage scheme capabilities.</p>
Index Termsfast Fourier transform; dynamic storage schemes; parallel memory performance; vectoraccesses; block accesses; constant-geometry FFT accesses; linear addresstransformations; XOR schemes; analytical results; quantitative analysis; bufferingeffects; pipelined memory systems; conflict-free access; memory bank cycle time; fastFourier transforms; memory architecture
D.T. Harper, III, "Block, Multistride Vector, and FFT Accesses in Parallel Memory Systems", IEEE Transactions on Parallel & Distributed Systems, vol. 2, no. , pp. 43-51, January 1991, doi:10.1109/71.80188
103 ms
(Ver 3.3 (11022016))