Path: csiph.com!usenet.blueworldhosting.com!feeder01.blueworldhosting.com!news.glorb.com!usenet.stanford.edu!not-for-mail
From: Juergen Sauermann <juergen.sauermann@t-online.de>
Newsgroups: gnu.g++.bug
Subject: Error Report
Date: Sat, 16 Apr 2016 18:05:12 +0200
Lines: 48
Approved: bug-gplusplus@gnu.org
Message-ID: <mailman.536.1460830706.7477.bug-gplusplus@gnu.org>
NNTP-Posting-Host: lists.gnu.org
Mime-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: quoted-printable
X-Trace: usenet.stanford.edu 1460830706 8196 208.118.235.17 (16 Apr 2016 18:18:26 GMT)
X-Complaints-To: action@cs.stanford.edu
To: bug-gplusplus@gnu.org
Envelope-to: bug-gplusplus@gnu.org
User-Agent: Mozilla/5.0 (X11; Linux i686; rv:31.0) Gecko/20100101 Thunderbird/31.4.0
X-ID: XZJ9hrZSoh-pLtbyTPsRPcCigTx2Nv-4ut+u6cxazmJ5mNgRztgkUpsG8kueaJGZs+
X-TOI-MSGID: 704596cb-0579-4611-b5e5-dd98391e3ceb
X-detected-operating-system: by eggs.gnu.org: GNU/Linux 3.x
X-Received-From: 194.25.134.20
X-Mailman-Approved-At: Sat, 16 Apr 2016 14:18:25 -0400
X-BeenThere: bug-gplusplus@gnu.org
X-Mailman-Version: 2.1.21
Precedence: list
List-Id: Bug reports for the GNU C++ compiler <bug-gplusplus.gnu.org>
List-Unsubscribe: <https://lists.gnu.org/mailman/options/bug-gplusplus>, <mailto:bug-gplusplus-request@gnu.org?subject=unsubscribe>
List-Archive: <http://lists.gnu.org/archive/html/bug-gplusplus/>
List-Post: <mailto:bug-gplusplus@gnu.org>
List-Help: <mailto:bug-gplusplus-request@gnu.org?subject=help>
List-Subscribe: <https://lists.gnu.org/mailman/listinfo/bug-gplusplus>, <mailto:bug-gplusplus-request@gnu.org?subject=subscribe>
Xref: csiph.com gnu.g++.bug:1

Hi,

I habe received the error report below on the bug-apl mailing list.

But the problem seems to be with the g++ compiler.

The effect is that -8 to the 1/3-power (=3D cubic root of -8) gives the
correct result when compiled with -O2, but 0 with -O5.

See http://lists.gnu.org/archive/html/bug-apl/2016-04/index.html for the=20
entire thread.

Best Regards,
J=C3=BCrgen Sauermann

> On Apr 13, 2016, at 7:03 PM, Kacper Gutowski<mwgamera@gmail.com>  wrote=
:
>
> On Thu, Apr 14, 2016 at 1:54 AM, Xiao-Yong Jin wrote:
>> Exactly so.  =E2=80=98-Ofast -fno-finite-math-only=E2=80=99 gives corr=
ect results.
>> What=E2=80=99s going on here?  I don=E2=80=99t expect this has anythin=
g to do with NaN or Inf.
> pow(-8., 1./3) is nan.
> And -ffinite-math-only eliminates isfinite check at FloatCell.cc:422.

That explains it.  Thanks.

Here is some amusement:

g++ -Ofast -fno-unsafe-math-optimizations:
     pow(-8., 1./3.)  =3D>  nan
     isfinite         =3D>  1

g++ -Ofast -fno-unsafe-math-optimizations -fno-finite-math-only:
     pow(-8., 1./3.)  =3D>  nan
     isfinite         =3D>  0

g++ -Ofast -fno-finite-math-only:
     pow(-8., 1./3.)  =3D>  -2
     isfinite         =3D>  1

clang++ -march=3Dnative -Ofast:
     pow(-8., 1./3.)  =3D>  nan
     isfinite         =3D>  0