fernau%ira.uka.de@RELAY.CS.NET (Henning Fernau) (05/29/90)
\I haven't read the following reference, but it sounds good: Q.F. Stout: Pyramids and Hypercubes, IN: V. Cantoni and S. Levialdi (Eds.): Pyramidal Systems for Computer Vision. Springer-Verlag, New York, 1986, 75--89. Yours sincerely, Henning Fernau Lehrstuhl Informatik fuer Ingenieure und Naturwissenschaftler University of Karlsruhe Am Fasanengarten 5 D-7500 Karlsruhe 1 phone: [West Germany](0721)608-4336 email:fernau@ira.uka.de
muttiah@cs.purdue.edu (Ranjan Samuel Muttiah) (05/30/90)
In article <9143@hubcap.clemson.edu> fernau%ira.uka.de@RELAY.CS.NET (Henning Fernau) writes: >\I haven't read the following reference, but it sounds good: > >Q.F. Stout: Pyramids and Hypercubes, IN: V. Cantoni and S. Levialdi (Eds.): >Pyramidal Systems for Computer Vision. Springer-Verlag, New York, 1986, >75--89. Yes, that was the one I was interested in. Thank everyone who sent in their help. I was _speculating_ on using a pyramidal system for something like a Fukushima net. If I may summarize (and please do feel free to correct me): For nodes that are a maximum of three links apart it is just a matter of overlaying so to speak the traditional gray codes for meshes. Each sub block of the overlay has a specific parent node responsible for the kids below. For a maximum of two links apart Stout gives a formula without proof [though I'm too lazy to prove it, induction should work I think]. The formula does seem valid. One should note that in a pyramid there can be an odd number of communication links for messages that are sent and received back to the same place where as the hypercube only has an even number IF one uses only the required number of processors. There is always the case of using a "dummy" processor for communication purposes only! Problems:Pyramids seem to be rather inefficient for interlevel computations. The more one does computations at _each_ level the more efficient the machine becomes.