Path: csiph.com!v102.xanadu-bbs.net!xanadu-bbs.net!news.glorb.com!solaris.cc.vt.edu!news.vt.edu!newsfeed-00.mathworks.com!panix!not-for-mail
From: Grant Edwards <invalid@invalid.invalid>
Newsgroups: comp.os.linux.development.apps
Subject: Re: cast double to short problem
Date: Wed, 7 Mar 2012 15:23:39 +0000 (UTC)
Organization: PANIX Public Access Internet and UNIX, NYC
Lines: 29
Message-ID: <jj7uhr$prg$1@reader1.panix.com>
References: <jj64g2$fb8$1@dont-email.me> <jj6650$fa0$1@reader1.panix.com> <jj6714$tbh$1@dont-email.me> <jj6a7s$l7k$1@reader1.panix.com> <jj7cii$ahl$1@dont-email.me>
NNTP-Posting-Host: dsl.comtrol.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Trace: reader1.panix.com 1331133819 26480 64.122.56.22 (7 Mar 2012 15:23:39 GMT)
X-Complaints-To: abuse@panix.com
NNTP-Posting-Date: Wed, 7 Mar 2012 15:23:39 +0000 (UTC)
User-Agent: slrn/pre0.9.9-102 (Linux)
Xref: csiph.com comp.os.linux.development.apps:449

On 2012-03-07, Noob <root@127.0.0.1> wrote:
> Grant Edwards wrote:
>
>> $ man round
>> #include <math.h>
>> i = (short) round(MyDoubleValue * 100.0);
>
> Depending on so-called rounding modes, some CPUs do not round as
> mathematicians expect.

IMO, depending on 1.5 or 2.5 being rounded one way or the other is a
bug, but I realize that in some statistical applications it matters.

However, the CPU type should have nothing to do with it -- the
behavior of the libm round() function is well defined to "round
halfway cases away from zero".  If you want half-way values to round
to nearest even then the round() man page says that you should call
rint().

> See, for example, the "nearest even" mode where both 1.5 and 2.5
> (which can be represented exactly in binary FP) are rounded to 2.
>
> http://www.ocf.berkeley.edu/~horie/rounding.html


-- 
Grant Edwards               grant.b.edwards        Yow! PARDON me, am I
                                  at               speaking ENGLISH?
                              gmail.com