Path: utzoo!utgpu!water!watmath!clyde!rutgers!mcnc!gatech!purdue!i.cc.purdue.edu!k.cc.purdue.edu!l.cc.purdue.edu!cik
From: cik@l.cc.purdue.edu (Herman Rubin)
Newsgroups: comp.arch
Subject: Re: Performance increase - a suggestion
Summary: And what if you need more?
Message-ID: <671@l.cc.purdue.edu>
Date: 1 Feb 88 12:09:45 GMT
References: <235@unicom.UUCP> <9375@steinmetz.steinmetz.UUCP>
Organization: Purdue University Statistics Department
Lines: 20


Whatever accuracy is built into the floating point processor, more may
be needed.  If the appropriate hardware is included, almost double of 
the "usual" is relatively easily obtainable, but no more.  For more, it
is necessary to go to integer arithmetic--even when there is no integer
arithmetic, it is necessary to force the floating point hardware to 
treat integers.

Thus if more accuracy than anticipated is needed, good integer arithmetic
is called for.  This means unsigned multiplication and division, among
other not necessarily available constructs.  I suggest that however many
bits in the mantissa of floating point numbers are directly treatable by
the hardware, that integer multiplication of numbers of that accuracy ,
with the product of twice the length, be hardware, and that this be
unsigned, and additional reasonable hardware to facilitate the arithmetic
be available.
-- 
Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907
Phone: (317)494-6054
hrubin@l.cc.purdue.edu (ARPA or UUCP) or hrubin@purccvm.bitnet