[Insight-developers] Define _GLIBCXX_DEBUG on dashboard?

[Bradley Lowekamp]

Sean,

I glances at some of those logs.

Here is a bit from one of the transform tests longs:


Exception Type:  EXC_BAD_ACCESS (SIGSEGV)
Exception Codes: EXC_I386_GPFLT

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   ITKTransformTestDriver        	0x0000000103757d91 vnl_sse<double>::matrix_x_vector(double const*, double const*, double*, unsigned int, unsigned int) + 337 (vnl_sse.h:482)
1   ITKTransformTestDriver        	0x0000000103757c30 vnl_vector<double>::vnl_vector(vnl_matrix<double> const&, vnl_vector<double> const&, vnl_tag_mul) + 272 (vnl_vector.h:247)
2   ITKTransformTestDriver        	0x0000000103757b15 vnl_vector<double>::vnl_vector(vnl_matrix<double> const&, vnl_vector<double> const&, vnl_tag_mul) + 37 (vnl_vector.h:247)
3   ITKTransformTestDriver        	0x0000000103114e04 vnl_vector<double> operator*<double>(vnl_matrix<double> const&, vnl_vector<double> const&) + 36 (vnl_vector.h:503)
4   ITKTransformTestDriver        	0x0000000103126710 vnl_vector<double> operator*<double, 3u, 3u>(vnl_matrix_fixed<double, 3u, 3u> const&, vnl_vector<double> const&) + 64 (vnl_matrix_fixed.h:1013)
5   ITKTransformTestDriver        	0x0000000103126694 itk::Matrix<double, 3u, 3u>::operator*(vnl_vector<double> const&) const + 36 (itkMatrix.hxx:220)
6   ITKTransformTestDriver        	0x000000010312144e itk::MatrixOffsetTransformBase<double, 3u, 3u>::TransformVector(vnl_vector_fixed<double, 3u> const&) const + 78 (itkMatrixOffsetTransformBase.hxx:209)
7   ITKTransformTestDriver        	0x000000010326404b itkVersorTransformTest(int, char**) + 4555 (itkVersorTransformTest.cxx:244)
8   ITKTransformTestDriver        	0x0000000102fabc63 main + 1347 (ITKTransformTestDriver.cxx:332)
9   libdyld.dylib                 	0x00007fff8973c7e1 start + 1

https://github.com/Kitware/ITK/blob/master/Modules/ThirdParty/VNL/src/vxl/core/vnl/vnl_sse.h#L482

And you are turning on the VNL SSE CMake flag, which I guess you are unique in doing. I don't see any effort in VNL/ITK to make the doubles aligned at all... I wonder if that is the problem.

I bet all those failing transform test would past on the system if this option was disabled. Here is the CMake documentation for the option: 

VNL_CONFIG_ENABLE_SSE2: Enable Streaming SIMD Extensions 2 optimisations (hardware dependant). Currently broken. For use by VNL developers only.

I have my doubts that we should even be displaying this as a top level CMake option.

Brad

On May 29, 2013, at 11:20 AM, Sean McBride <sean at rogue-research.com> wrote:

> On Wed, 29 May 2013 09:04:31 -0400, Bradley Lowekamp said:
> 
>> I am currently a little fired up to tack down more bugs.
>> 
>> Do you think you could configure a torture build but turn on only the
>> Core modules? 
> 
> Brad,
> 
> Yup, I'll add flags slowly, using my VTK experience as a guide.  Last night added 15 new failures with just 4 ubsan flags (-fsanitize=bool,return,integer-divide-by-zero,shift).  The flags are described here:
> 
> <http://clang.llvm.org/docs/UsersManual.html#controlling-code-generation>
> 
> the ones marked "ILLEGAL" are the ones that crash due to ubsan generating an illegal instruction:
> 
> <http://open.cdash.org/viewTest.php?onlyfailed&buildid=2921264>
> 
> Alas, the dashboard does not show the backtrace, but looking at them shows many failures look to be ultimately duplicates.  I'll send you the crash logs offlist.
> 
> Cheers,
> 
> -- 
> ____________________________________________________________
> Sean McBride, B. Eng                 sean at rogue-research.com
> Rogue Research                        www.rogue-research.com 
> Mac Software Developer              Montréal, Québec, Canada
> 
>