This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: gsl and Intel C++ compiler for Linux
- From: John Ketchum <johnk at qualcomm dot com>
- To: Brian Gough <bjg at network-theory dot co dot uk>
- Cc: gsl-discuss at sources dot redhat dot com
- Date: Wed, 08 May 2002 17:39:20 -0400
- Subject: Re: gsl and Intel C++ compiler for Linux
- References: <5.1.0.14.0.20020508103009.03d2ec90@mail1.qualcomm.com><5.1.0.14.0.20020503151745.01df6f00@mail1.qualcomm.com><5.1.0.14.0.20020508103009.03d2ec90@mail1.qualcomm.com>
I found a compiler switch (-mp) that corrects this problem. There are
several switches that control accuracy in somewhat obscure ways. I have
gotten to the point now where make check fails in specfunc, and I haven't
been able to find a combination of compiler switches that corrects the
problem, which appears to be an accuracy problem with exponential functions
and legendre functions.
The compiler is supposed to be gcc-compatible. But so far, it is not doing
very well.
At 10:14 PM 5/8/2002 +0100, Brian Gough wrote:
>John Ketchum writes:
> > FAIL: double x = DBL_MIN/2^1, mantissa
> > (0000000000000000000000000000000000000000000000000000 observed vs
> > 1000000000000000000000000000000000000000000000000000 expected)
> > FAIL: double x = DBL_MIN/2^1, type is DENORMAL (5 observed vs 4 expected)
>
>Seems that it rounds denormalized numbers to zero.
>
>Brian
John Ketchum
Qualcomm Inc.
9 Damonmill Square Suite 2A
Concord, MA 01742
Office: 60915
Direct: 781-276-0915
Fax: 781-276-0901
Mobile: 617-312-7023