[Insight-developers] level set/vnl problems on SGI: update

Lorensen, William E (Research) lorensen@crd.ge.com
Tue, 15 Apr 2003 16:53:21 -0400


Yes. SGI's assume that char is "unsigned". We should always specified signed
if we want it. I though I had caught them all.


-----Original Message-----
From: Lydia Ng [mailto:lng@insightful.com]
Sent: Tuesday, April 15, 2003 4:03 PM
To: insight-developers@public.kitware.com
Subject: [Insight-developers] level set/vnl problems on SGI: update


I believe the level set problems on SGI are now fixed.

It is related to "char/signed char" issue and the fixes made to
itkNumericsTraits.h on Mar 18th.

The SparseFieldLevelSetImageFilter uses a "char" image to store status
information. I've make it explicitly "signed char" and the tests + the
regression tests are now passing on the uiowa SGI machines.

The test now passes on the GE SGI's. However, the regression tests seem
to be still failing. Perhaps they regression tests are not rebuilding
properly?

- Lydia
 

> -----Original Message-----
> From: Bill Hoffman [mailto:bill.hoffman@kitware.com]
> Sent: Tuesday, April 08, 2003 7:17 AM
> To: Miller, James V (Research); Lydia Ng; insight-
> developers@public.kitware.com
> Subject: RE: [Insight-developers] level set/vnl problems on SGI?
> 
> It looks like they are doing the right thing:
> 
> vcl/sgi/vcl_cmath.h
> inline float       vcl_sqrt(float  x) { return ::sqrtf(x); }
> inline double      vcl_sqrt(double x) { return ::sqrt(x); }
> inline long double vcl_sqrt(long double x) { return ::sqrtl(x); }
> 
> inline float       vcl_abs (float       x) { return ::fabsf(x); }
> inline double      vcl_abs (double      x) { return ::fabs(x); }
> inline long double vcl_abs (long double x) { return ::fabsl(x); }
> 
> However, perhaps this is the problem.   Perhaps the code is
> depending on a promotion to double for some of the operations,
> and the SGI is using float versions.   sqrt and abs are the only
> ones with the problem.
> 
> 
> 
> -Bill
> 
> 
> At 10:03 AM 4/8/2003, Miller, James V (Research) wrote:
> >I believe the SGI provides overloaded versions of
> >abs and sqrt for various data types.
> >
> >I would guess that these vnl functions are not being
> >properly defined on the SGI.
> >
> >Jim
> >
> >
> >> -----Original Message-----
> >> From: Bill Hoffman [mailto:bill.hoffman@kitware.com]
> >> Sent: Monday, April 07, 2003 3:09 PM
> >> To: Lydia Ng; insight-developers@public.kitware.com
> >> Subject: Re: [Insight-developers] level set/vnl problems on SGI?
> >>
> >>
> >> There are no known issues with these functions, and
> >> they are being used in other places that seem to work.
> >> However, it sounds like you are close to finding the problem.
> >> I would suggest adding some more prints, so we can find
> >> out exactly where things start to differ, and which of the
> >> functions is causing the trouble.
> >>
> >> -Bill
> >>
> >>
> >>
> >>
> >> >I think March 17th was around the time the VNL stuff was
> >> being changed
> >> >over.
> >> >The LevelSetFunction uses the following vnl/vcl functions:
> >> >
> >> > vnl_math_abs
> >> > vnl_math_min
> >> > vnl_math_max
> >> > vcl_sqrt
> >> >
> >> >
> >> >Are there any known issues with any of these functions on SGI?
> >> >
> >> >
> >> >- Lydia
> >> >
> >>
> >>
> >>
> >> _______________________________________________
> >> Insight-developers mailing list
> >> Insight-developers@public.kitware.com
> >> http://public.kitware.com/mailman/listinfo/insight-developers
> >>
> >_______________________________________________
> >Insight-developers mailing list
> >Insight-developers@public.kitware.com
> >http://public.kitware.com/mailman/listinfo/insight-developers
> 
> 

_______________________________________________
Insight-developers mailing list
Insight-developers@public.kitware.com
http://public.kitware.com/mailman/listinfo/insight-developers