[Insight-developers] Rounding functions in itkMacro.h

[Tom Vercauteren]

Hi Wes,

Sorry if I wasn't clear enough. I anticipate this patch to fix the
nightlies of RogueResearch3 and RogueResearch5 but not Dash14 which I
believe has nothing to do with the rounding functions*.

I also anticipate this patch to fix Kent's issue:
http://www.itk.org/mailman/private/insight-developers/2009-May/012545.html

Tom

* As mentioned by Bill, Dash14 has been failing before the rounding
functions were checked in:
http://www.itk.org/mailman/private/insight-developers/2009-May/012559.html

On Tue, May 19, 2009 at 15:56, Wes Turner <wes.turner at kitware.com> wrote:
> Hey Tom,
> Just to be clear ... do you anticipate this fixing the Nightlies for:
> RogueResearch3 (50 Build Errors) ?
> RogueResearch5 (50 Build Errors) ?
> Dash14               (50 Build Errors) ?
> I'm trying to focus my effort on what remains, and would hate to chase an
> issue that you've already resolved!
> - Wes
>
> On Tue, May 19, 2009 at 9:44 AM, Tom Vercauteren <tom.vercauteren at m4x.org>
> wrote:
>>
>> Hi, all
>>
>> I have just committed a patch to vnl_math.h:
>>
>>  http://www.itk.org/cgi-bin/viewcvs.cgi/Utilities/vxl/core/vnl/vnl_math.h?root=Insight&r1=1.10&r2=1.8&sortby=date
>>
>> It should fix the wrapping and ppc64 builds failures.
>>
>> Tom
>>
>> On Tue, May 19, 2009 at 00:17, Michel Audette
>> <michel.audette at kitware.com> wrote:
>> > Hi Tom,
>> >
>> > This patch seems to have improved things. The number of failures related
>> > to
>> > this flag and the centered coordinates is down to 16 on my 64-bit Linux
>> > machine.
>> >
>> > Best wishes,
>> >
>> > Michel
>> >
>> > On Mon, May 18, 2009 at 3:10 PM, Tom Vercauteren
>> > <tom.vercauteren at m4x.org>
>> > wrote:
>> >>
>> >> Just a wild guess since I have never played with wrapping before.
>> >> Could the error in:
>> >>  http://www.cdash.org/CDash/viewBuildError.php?buildid=335436
>> >> come from gcc-xml and not from visual c++? Are wrapping builds mixing
>> >> the use of the native compiler and gcc-xml?
>> >>
>> >> In that case it might be that the native compiler allowed for the use
>> >> of sse2 xml. Hence VNL_CONFIG_ENABLE_SSE2_ROUNDING would be on even
>> >> for the gcc-xml build that might not support it.
>> >>
>> >> In any case, would there be a reson for gcc-xml to use the optimized
>> >> vnl_rnd functions? We might as well fallback to the vanilla
>> >> implementation when gcc-xml is used.
>> >>
>> >> If this sounds reasonnable, one potential patch would be to change in
>> >> vnl_math_rnd.h
>> >>  #if VNL_CONFIG_ENABLE_SSE2_ROUNDING
>> >> by something like (needs syntax checking)
>> >>  #if VNL_CONFIG_ENABLE_SSE2_ROUNDING && (!defined(__GCC_XML__))
>> >> or better by changing
>> >>
>> >> #if VNL_CONFIG_ENABLE_SSE2_ROUNDING
>> >> # if !VXL_HAS_EMMINTRIN_H
>> >> #   error "Required file emmintrin.h for SSE2 not found"
>> >> # else
>> >> #   include <emmintrin.h> // sse 2 intrinsics
>> >> # endif
>> >> #endif
>> >>
>> >> into
>> >>
>> >> #if VNL_CONFIG_ENABLE_SSE2_ROUNDING && (!defined(__GCC_XML__))
>> >> # if !VXL_HAS_EMMINTRIN_H
>> >> #   error "Required file emmintrin.h for SSE2 not found"
>> >> # else
>> >> #   include <emmintrin.h> // sse 2 intrinsics
>> >> #   define USE_SSE2_IMPL 1
>> >> # endif
>> >> #endif
>> >>
>> >> Then use USE_SSE2_IMPL in lieu of VNL_CONFIG_ENABLE_SSE2_ROUNDING in
>> >> that
>> >> file.
>> >>
>> >>
>> >> Also it needs changing
>> >>  #if defined(__GNUC__) && (!defined(__APPLE__)  || !defined(__ppc__) )
>> >> by something like
>> >>  #if defined(__GNUC__) && (!defined(__ppc__))  &&
>> >> (!defined(__ppc64__)) && (!defined(__GCC_XML__))
>> >>
>> >> Does that make sense to you?
>> >> Tom
>> >>
>> >> On Mon, May 18, 2009 at 20:20, Wes Turner <wes.turner at kitware.com>
>> >> wrote:
>> >> > Tom,
>> >> >
>> >> > I am open to suggestions.  There is a bug in one of the ImageReader
>> >> > tests that is causing some of the Windows and Mac platforms to wait
>> >> > for a timeout when running itkImageFileReaderDimensionsTest_MHD.  I
>> >> > am
>> >> > hoping that if I can decipher this, the missing Nightly dashboards
>> >> > will begin showing up as of tomorrow.  Until then (and after then)
>> >> > solutions to this problem are welcome!
>> >> >
>> >> > - Wes
>> >> >
>> >> > (There is another set of failures on the DASH14 Nightly that I need
>> >> > to
>> >> > re  neesolve as   we
>> >> > On Mon, May 18, 2009 at 1:32 PM, Tom Vercauteren
>> >> > <tom.vercauteren at m4x.org> wrote:
>> >> >> Hi,
>> >> >>
>> >> >> The dashboard is showing a mysterious failure:
>> >> >> http://www.cdash.org/CDash/viewBuildError.php?buildid=335436
>> >> >>
>> >> >> Apparently on this MSVC 7 build, __intXX is not defined, e.g.:
>> >> >>  '__int32' does not name a type
>> >> >>
>> >> >> However, the __intXX types are listed to be fundamental types for
>> >> >> this
>> >> >> compiler:
>> >> >>  http://msdn.microsoft.com/en-us/library/29dh1w7z(VS.71).aspx
>> >> >>
>> >> >> Also we are not even directly using these types. They are apparently
>> >> >> pulled in by emmintrin.h. This build seem to have wrapping turned
>> >> >> on.
>> >> >> Could this be somehow related to the error that Kent Williams just
>> >> >> reported:
>> >> >>
>> >> >>
>> >> >>  http://www.itk.org/mailman/private/insight-developers/2009-May/012545.html
>> >> >>
>> >> >> This is beyond my understanding. Any thoughts?
>> >> >>
>> >> >> Tom
>
>
>
> --
> Wesley D. Turner, Ph.D.
> Kitware, Inc.
> R&D Engineer
> 28 Corporate Drive
> Clifton Park, NY 12065-8662
> Phone: 518-371-3971 x120
>