[Insight-developers] test failing on icc

[Hans Johnson]

Bill,

After the TCON request yesterday, I have renamed all the uiowa build names.
This is going to cause 8 platforms to not be submitted, and 8 "new"
platforms to be added.  This is just going to make the dashboard ugly.  If
the names need more modification, just let me know.

In addition, I will remove any regression tests that you want.   There are
no duplicates, but I always do 2 compilations on each platform: a debug
version with "-g" flags and full warnings, and a fast version with "-O3"
flags and optimizations. It is my experience that differences between these
two build types often flush out subtle coding errors or blemishes.  I have
found it really useful to find missing parenthesis in mathematical
expressions when a 64bit SGI and 32bit linux in debug give one answer, and
the fast versions give the same answers, but the debug vs. fast versions
give different answers.

To go even further, I have used the ICC compiler in debug, fast, and a
version that relaxes all IEEE floating point requirements for super fast,
but non-portable, execution.  This has helped us find the parts of our code
that are very sensitive to floating point precision, and we were able to add
some parenthesis, and reorder some loops to make the algorithm more stable.

Regards,
Hans





-----Original Message-----
From: insight-developers-admin at itk . org
[mailto:insight-developers-admin at itk . org]On Behalf Of Bill Hoffman
Sent: Friday, August 08, 2003 12:24 PM
To: C. Aaron Cois; insight-developers at public . kitware . com
Subject: Re: [Insight-developers] test failing on icc


Contact the person running the dashboard and have them run a debugger.

-Bill


At 02:18 PM 8/8/2003, C. Aaron Cois wrote:
>Our ./Testing/Code/Common/itkMedialNodeCorrespondencesTest is passing on
everything but icc compilers (which unfortunately are numerous)  Reports
indicating that it is hitting a seg fault relatively early in execution
(judging from the run time), but I can't tell much more than that.  I don't
suppose any developers familiar with icc could offer some advice?
>
>C. Aaron Cois
>Graduate Student
>Dept. of Bioengineering
>Dept. of Computer Science
>University of Pittsburgh



_______________________________________________
Insight-developers mailing list
Insight-developers at itk . org
http://www . itk . org/mailman/listinfo/insight-developers