Path: utzoo!utgpu!water!watmath!clyde!bellcore!decvax!decwrl!ucbvax!ucsd!ucsdhub!esosun!seismo!uunet!mcvax!dik
From: dik@cwi.nl (Dik T. Winter)
Newsgroups: comp.arch
Subject: Re: Vectorising conditional code.
Message-ID: <7585@boring.cwi.nl>
Date: 18 Jul 88 01:28:03 GMT
References: <893@garth.UUCP> <28200176@urbsdc> <963@garth.UUCP> <7584@boring.cwi.nl> <837@l.cc.purdue.edu>
Organization: CWI, Amsterdam
Lines: 18

In article <837@l.cc.purdue.edu> cik@l.cc.purdue.edu (Herman Rubin) writes:
 > For example, suppose that the present procedure to deal with double precision
 > comples numbers is to have a vector DC in which each item is four machine
 > words, and the vector operations can do such things as compute both the
 > upper and lower parts of a double precision operation on two words.  Then
 > we should have four vectors, say RU (the upper part of the real part), RL,
 > CU, and CL.  This way it would not be necessary to set up the scratch vectors 
 > and to start out by separating the DC vector into the four parts, which we
 > are likely to have to do anyhow!
 > 
 > Other than the changes in the languages needed to accomplish this, I see little
 > problem except for input and output, and possibly the concatenation of vectors.
A problem arises if we want to pass an element as a variable to a subroutine.
(Things like this were possible in Algol 68 anyway.)
-- 
dik t. winter, cwi, amsterdam, nederland
INTERNET   : dik@cwi.nl
BITNET/EARN: dik@mcvax