NetNews Usenet Archive 1992 #20/NN_1992_20.iso/spool/comp/sys/transput/995

Text File | 1992-09-08 | 1.7 KB | 34 lines

Newsgroups: comp.sys.transputer
Path: sparky!uunet!haven.umd.edu!darwin.sura.net!Sirius.dfn.de!rz.ruhr-uni-bochum.de!jan
From: [e-mail] (Jan Vorbrueggen)
Subject: Re: Protocol query
Message-ID: <[e-mail]>
In-reply-to: [e-mail]'s message of 4 Sep 92 12:51:00 GMT
Organization: Inst. f. Neuroinformatik, Ruhr-Universitaet Bochum, FRG
References: <[e-mail]>
Date: 7 Sep 92 14:28:20
Lines: 22

What you're really trying to do is extracting a part of memory with
nonunit stride. (Unit stride is of course handled just by using
assignment in occam and translated into the MOVE instruction.) The
T800 and followon (T425, T400, T805, T801) have MOVE2D exactly for
this purpose. Unfortunately, the compiler's builtin function has
a braindamaged interface (IMHO) requiring massive RETYPES, as well
as some code generation bugs (if certain arguments aren't constant
or VALs, space is allocated for their computation, but the computation
is never done :)). So I defined my own routine(s) for the purpose.
It is very efficient: For instance, if you're doing, say, a 2D FFT,
for the column part it is better to extract each column into a 1D array,
perform the FFT on it, and reinsert it with MOVE2D. This works because
you save massively on address computations, need only a 1D FFT that is
unaware it is working on a 2D array, and you can PLACE the temporary
array into the WORKSPACE.

So the solution is to extract your data from your 2D array into a buffer
using MOVE2D, send that, and scatter it at the receiving end via MOVE2D,
if required.

 Jan Vorbrueggen, Inst. f. Neuroinformatik, RuhrUniversitaet Bochum, FRG
 [e-mail]