NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c++-part2/10114

Text File | 1996-08-05 | 1.9 KB | 44 lines

Path: mozo.cc.purdue.edu!not-for-mail
From: [e-mail] (Herman Rubin)
Newsgroups: comp.arch.arithmetic,comp.lang.c,comp.lang.c++
Subject: Re: Access carry flag from C
Date: 6 Mar 1996 04:34:29 -0500
Organization: Purdue University Statistics Department
Message-ID: <4hjm75$[e-mail]>
References: <[e-mail]> <[e-mail]> <4hhbt8$[e-mail]> <[e-mail]>
NNTP-Posting-Host: b.stat.purdue.edu

In article <[e-mail]>, Dik T. Winter <[e-mail]> wrote:
>In article <4hhbt8$[e-mail]> [e-mail] (Herman Rubin) writes:
> > In article <[e-mail]>,
> > Franτois Grieu <[e-mail]> wrote:

> > >   if (x + yl < yl) ++yh; /* test for overflow on next line */
> > >   yl += x;
>...
> > Notice that in the above code, each addition must be performed twice.

>Wrong.  Even a mediocre compiler will optimize this away.

Even if this is done, there are still two additions, as the comparison
is normally done by an addition in hardware.  Add this to the overhead
of the test, and much time is used.

> > BTW, there is a weakness in the code; ++yh can cause a carry, so
> > another test is needed.  As testing is expensive, what is needed in
> > this case is not a test for carry, but the ability to add the carry
> > bit to something.

>But the code above also works on machines without carry bit.

This still affects the result.

On machines without the carry bit, an efficient mp routine would
use a few less bits in the word, so that all these tests would
not be needed.  One is rarely interested in EXACTLY double the
precision provided by the hardware, just as one is rarely interested
in exactly the precision of the hardware.
--
Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
[e-mail]     Phone: [phone]    FAX: [phone]