Hi, thanks to everybody for your useful hints. I solved my problem looking on this particular error on intel compiler forum. I need to add to my original configure this:
—CXXFLAGS=‘-openmp -parallel’ because i use this option for C and Fortran Intel compiler too. Using this option (-openmp), intel compiler links the thread safe version of Intel(R) MPI library. If i use these thread safe library for C and Fortran, i must use also thread safe library for C++ if i want to avoid during compilation these type of errors. Once i added, this flag everything went fine, and i was able also to successful do a ‘make’ and a ‘test’. I hope that this could be helpful to someone else. Thanks again for your time Pierpaolo Il giorno 01/apr/2014, alle ore 18:39, Satish Balay <[email protected]> ha scritto: > On Tue, 1 Apr 2014, Pierpaolo Minelli wrote: > >> I am a fortran programmer so, i don’t know how to check this. >> I have the latest suite of Intel Compiler. (Intel Cluster Studio XE 2013). >> Which type of check can i do to see if it is working c++ compiler? > > The error is from intel MPI [not necessarily the C++ compiler] > >> In the configure.log, it seems that some sources are compiled while this no, >> or i have seen wrong? > > Its possible that intel compilers/mpi should be used in a perticular > way - otherwise things break. > > However I see you are using: > > Configure Options: --configModules=PETSc.Configure > --optionsModule=PETSc.compilerOptions > --PETSC_DIR=/raid0/users/cscppm59/Installers/petsc --PETSC_ARCH=petsc_dev > --with-debugger=idbc --with-cc=mpiicc --with-cxx=mpiicpc --with-fc=mpiifort > --CFLAGS="-openmp -parallel" --FFLAGS="-ftz -O3 -prec-div -xSSE2 -mcmodel > medium -shared-intel -openmp -parallel -fpp" --with-large-file-io=1 > --with-debugging=0 --with-mpiexec="mpirun -f ~/mpd.hosts -r ssh" > --with-blas-lapack-dir=/opt/intel/mkl/lib/intel64/ --with-threadcomm > --with-openmp --with-x=0 --download-superlu_dist --download-parmetis > -download-hypre -download-ml --download-metis > > Do you need all the options? Esp -with-threadcomm --with-openmp > CFLAGS="-openmp -parallel" etc? Or is it that you assumed > this is the way to get max performance off these MPI compilers? > > Unless you really need these features - I'll suggest removing them. > > also you can try the addtional options - and see if it makes a difference. > > --with-clib-autodetect=0 --with-fortranlib-autodetect=0 > --with-cxxlib-autodetect=0 LIBS="-Bstatic -lifcore -Bdynamic" > > However - if you are developing code - and not doing performance runs > [say on a cluster] - its best to build with gcc/gfortran and > --download-mpich. > > Satish
