Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!techunix.BITNET!dario
From: dario@techunix.BITNET (Dario Ringach)
Newsgroups: comp.sys.transputer
Subject: Configuring relaxation algorithms
Keywords: configuration
Message-ID: <7885@techunix.BITNET>
Date: 28 Feb 89 06:13:02 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: Technion - Israel Inst. Tech., Haifa Israel
Lines: 13

I suppose the most straightforward way to implement a relaxation labeling
algorithm (say for image processing) in OCCAM is by mapping a process to
each pixel which in each iteration receives the state of its 4-neighbors
and updates its own state.  Assume that I have an NxN grid so we actually
have N^2 OCCAM processes running in parallel.  Now, you are given an MxM torus
of Transputers (M<N) and you are asked to configure your software for such
an architecture..., impossible right?  Instead of the trivial solution of
writing the algorithm as MxM processes, each working in subpictures and
directly mapped onto the Tps., is there a way to write it such that it could
be possible to configure it for any M?  Why a Tp link implements only two OCCAM
channels and not several of them?  Thanks in advance!
--
Dario