compilation error in psfcorr when using icpc

[tomaszkacprzak]

I am running into this error when compiling with icpc.
gcc installation works fine, and I still get the API warnings.
Does it look familiar to you?

icpc -o pysrc/.obj/Angle.os -c -mkl -fPIC -Iinclude/galsim -Iinclude -I/home/kacprzak/local/include -I/home/kacprzak/local/tmv/include -I/home/kacprzak/local/include/python2.7 -I/home/kacprzak/local/python/2.7.3/lib/python2.7/site-packages/numpy/core/include pysrc/Angle.cpp
/home/kacprzak/local/python/2.7.3/lib/python2.7/site-packages/numpy/core/include/numpy/npy_deprecated_api.h(11): warning #1224: #warning directive: "Using deprecated NumPy API, disable it by #defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION"
  #warning "Using deprecated NumPy API, disable it by #defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION"
   ^

/usr/include/c++/4.4.6/cmath(528): error: identifier "__builtin_isnan" is undefined
        return __builtin_isnan(__type(__f));
               ^
          detected during:
            instantiation of "__gnu_cxx::__enable_if<std::__is_arithmetic<_Tp>::__value, int>::__type std::isnan(_Tp) [with _Tp=double]" at line 771 of "src/hsm/PSFCorr.cpp"
            instantiation of "void galsim::hsm::find_ellipmom_2(galsim::ConstImageView<T>, galsim::ConstImageView<int>, double &, double &, double &, double &, double &, double &, double &, double, int &) [with T=double]" at line 115 of "src/hsm/PSFCorr.cpp"

compilation aborted for src/hsm/PSFCorr.cpp (code 2)
scons: *** [src/hsm/.obj/PSFCorr.os] Error 2
scons: building terminated because of errors.

[rmjarvis]

I assume you mean on branch #343? This didn't get the fix from #298 yet. I just merged from master which includes that fix, so I think it should work now. For future reference, you can always check whether some branch has all the updates that are on master with:

git checkout master
git pull
git checkout some_other_branch
git pull
git merge master

If all is up to date, you should get:

$ git merge master
Already up-to-date.

I'm sure there is some git way to just check it it's consistent other than to try the merge, but since I always want to do the merge if it hasn't been done yet, this is my SOP.

[rmandelb]

Hi @tomaszkacprzak - just wanted to check whether this worked for you, in which case we will close this issue?

[tomaszkacprzak]

I tried to do compile it with the fix you mentioned Mike, but I am getting an error of other sort. I am pretty sure I messed up my intel-compiled TMV, Numpy or Boost, so I will start with redoing my intel version of GalSim and dependencies. Then I will report!

[rmandelb]

Hi @tomaszkacprzak - any update? I just noticed this hasn't been updated in 2 months.

[tomaszkacprzak]

Hi Rachel, sorry about that. I was using the gcc version successfully. I tried to recompile after updating galsim, but I ran into the same problem, but without the Numpy warning. It looks like boost linking errors..

icpc -o pysrc/.obj/Angle.os -c -O2 -msse2 -vec-report0 -fPIC -Iinclude/galsim -Iinclude -I/home/kacprzak/local/python/2.7.3/include -I/home/kacprzak/local/tmv-intel/include -I/home/kacprzak/local/include/python2.7 -I/home/kacprzak/local/python/2.7.3/lib/python2.7/site-packages/numpy/core/include pysrc/Angle.cpp
/usr/include/c++/4.4.6/cmath(528): error: identifier "__builtin_isnan" is undefined
        return __builtin_isnan(__type(__f));
               ^
          detected during:
            instantiation of "__gnu_cxx::__enable_if<std::__is_arithmetic<_Tp>::__value, int>::__type std::isnan(_Tp) [with _Tp=double]" at line 785 of "src/hsm/PSFCorr.cpp"
            instantiation of "void galsim::hsm::find_ellipmom_2(galsim::ConstImageView<T>, galsim::ConstImageView<int>, double &, double &, double &, double &, double &, double &, double &, double, int &, boost::shared_ptr<galsim::hsm::HSMParams>) [with T=double]" at line 130 of "src/hsm/PSFCorr.cpp"

compilation aborted for src/hsm/PSFCorr.cpp (code 2)
scons: *** [src/hsm/.obj/PSFCorr.os] Error 2
scons: building terminated because of errors.

There are errors in config.log:

ImportError: .sconf_temp/conftest_52_mod/check_bp.so: undefined symbol: _ZNK5boost6python7objects21py_function_impl_base9max_arityEv

icpc -o .sconf_temp/conftest_29 -openmp .sconf_temp/conftest_29.o -Llib -L/home/kacprzak/local/python/2.7.3/lib -L/home/kacprzak/local/tmv-intel/lib -lfftw3 -ltmv -lpthread
.sconf_temp/conftest_29.o: In function `tmv::SymMatrix<double, 0>::SymMatrix(long, double const&)':
.sconf_temp/conftest_29.cpp:(.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd[.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd]+0xfc8): undefined reference to `VTT for tmv::GenSymMatrix<double>'
.sconf_temp/conftest_29.cpp:(.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd[.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd]+0xfd5): undefined reference to `VTT for tmv::GenSymMatrix<double>'
.sconf_temp/conftest_29.cpp:(.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd[.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd]+0xfdf): undefined reference to `VTT for tmv::GenSymMatrix<double>'
.sconf_temp/conftest_29.cpp:(.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd[.gnu.linkonce.t._ZN3tmv9SymMatrixIdLi0EEC2ElRKd]+0xfea): undefined reference to `VTT for tmv::GenSymMatrix<double>'
.sconf_temp/conftest_29.o: In function `tmv::GenSymMatrix<double>::~GenSymMatrix()':
.sconf_temp/conftest_29.cpp:(.gnu.linkonce.t._ZN3tmv12GenSymMatrixIdED2Ev[.gnu.linkonce.t._ZN3tmv12GenSymMatrixIdED2Ev]+0x22): undefined reference to `VTT for tmv::GenSymMatrix<double>'
.sconf_temp/conftest_29.o:.sconf_temp/conftest_29.cpp:(.gnu.linkonce.t._ZTv0_n56_N3tmv12GenSymMatrixIdED0Ev[.gnu.linkonce.t._ZTv0_n56_N3tmv12GenSymMatrixIdED0Ev]+0x24): more undefined references to `VTT for tmv::GenSymMatrix<double>' follow
.sconf_temp/conftest_29.o: In function `tmv::SymMatrixView<std::complex<double>, 

And:

ImportError: .sconf_temp/conftest_50_mod/check_bp.so: undefined symbol: _ZNK5boost6python7objects21py_function_impl_base9max_arityEv

I can send the whole config.log if needed.

[rmjarvis]

I think the problem is that you are using TMV built with g++ rather than icpc. You need to build a version of TMV with icpc.

On my system where I test with both g++ and icpc, I use PREFIX = ~/icpc-install for icpc. (I also store the different gs_scons.conf files as gs_scons.icpc and gs_scons.g++ so I can quickly switch between them.) Anyway, I put the TMV library compiled with icpc in ~/icpc-install/lib.

[rmjarvis]

Actually, I take back my previous comment. I was basing this on the config.log stuff you posted, but rereading your post, I see that that's not relevant, since you did get past that to the .cpp compiling.

So, on to isnan. This is actually a big problem with the (pre-C++11) C++ standard. There was no standard library isnan function. It was defined as a macro in C99, but that isn't technically part of the standard C++. It's just a common extension that is often implemented by compilers. Sometimes as a macro. Sometimes as a function. So it looks like you've run into a conflict with it. (Now with C++11, they have finally adopted it as part of the standard, but we're not requiring C++11 for GalSim currently.)

For TMV, I rolled my own isnan function to avoid all the platform-dependent issues with it. But since we're using boost anyway, my recommendation is to use boost::math::isnan here. This seems to be the only place we use the isnan function, so it's not a big change.

[rmjarvis]

I went ahead and made the switch on master. Please see if that works for you.

[tomaszkacprzak]

Thanks! That did the trick, now it compiles and imports.