NRN_ENABLE_SHARED=OFF builds static library (for GPU support)

[pramodk]

• libnrnmech is now build as static library if NRN_ENABLE_SHARED=OFF
• this is required for GPU support with OpenACC because global
  variables from shared library are not initialized

Also related to issues mentioned in #646

[pramodk]

This now allows to embed GPU enabled CoreNEURON into NEURON as:

# build mechanisms for coreneuron
/gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/nrn/build_gpu_mod2c_python2/install/bin/nrnivmodl-core .

# link coreneuron with mechanisms into neuron
/gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/nrn/build_gpu_mod2c_python2/install/bin/nrnivmodl \
-loadflags \
"-acc -rdynamic -ldl -lrt \
-Wl,--whole-archive -Lx86_64/ -lcorenrnmech \
/gpfs/bbp.cscs.ch/ssd/apps/hpc/jenkins/deploy/external-libraries/2020-02-01/linux-rhel7-x86_64/gcc-8.3.0/cuda-10.1.243-aekzngafo3/lib64/libcudart_static.a \
-L/gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/nrn/build_gpu_mod2c_python2/install/lib \
-lcoreneuron -lcudacoreneuron  \
-Wl,--no-whole-archive \
-I/gpfs/bbp.cscs.ch/ssd/apps/hpc/jenkins/deploy/tools/2020-02-01/linux-rhel7-x86_64/gcc-8.3.0/hpe-mpi-2.21-7pbszh6v5u/include \
-L/gpfs/bbp.cscs.ch/ssd/apps/hpc/jenkins/deploy/tools/2020-02-01/linux-rhel7-x86_64/gcc-8.3.0/hpe-mpi-2.21-7pbszh6v5u/lib \
-lmpi++ -lmpi" .

cc: @iomaganaris

This is still not clean but now have all information required to implement clean solution (as separate PR).

[pramodk]

@nrnhines : was there any specific need to have libnrnmech.la target here?

[pramodk]

With this PR I have seen:

(cd .. ;  g++ -O2 -DVERSION_INFO='8.0.dev160' -std=c++11  -I /home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/include -I /home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/include/nrncvode -DAUTO_DLOPEN_NRNMECH=0 /home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/share/nrn/nrnmain.cpp -o x86_64/special \
  -L x86_64 -lnrnmech -L/home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/lib -lnrniv -lnrnmech -L/home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/lib -lnrniv -Wl,-rpath,'$ORIGIN' -Wl,-rpath,/home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/lib   )
/home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/lib/libnrniv.so: undefined reference to `pthread_create'
/home/vsts/work/1/s/nrn_test_venv_35/lib/python3.5/site-packages/neuron/.data/lib/libnrniv.so: undefined reference to `pthread_join'

and also below for static library build:

 => LINKING static library x86_64/libnrnmech.a Mod files: CaDynamics_E2.mod
 => LINKING executable x86_64/special LDFLAGS are:    -lreadline -lcurses -lform /gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/soft/install/linux-rhel7-x86_64/gcc-8.3.0/python-2.7.16-5x2gzf/lib/libpython2.7.so -Wl,-rpath,/gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/soft/install/linux-rhel7-x86_64/gcc-8.3.0/python-2.7.16-5x2gzf/lib -lpthread /gpfs/bbp.cscs.ch/ssd/apps/hpc/jenkins/deploy/tools/2020-02-01/linux-rhel7-x86_64/gcc-8.3.0/hpe-mpi-2.21-7pbszh6v5u/lib/libmpi.so -Wl,-rpath,/gpfs/bbp.cscs.ch/ssd/apps/hpc/jenkins/deploy/tools/2020-02-01/linux-rhel7-x86_64/gcc-8.3.0/hpe-mpi-2.21-7pbszh6v5u/lib
/gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/nrn/build_gpu_mod2c_python2/install/share/nrn/nrnmain.cpp:
/usr/bin/ld: /gpfs/bbp.cscs.ch/project/proj16/kumbhar/pramod_scratch/nrn/build_gpu_mod2c_python2/install/lib/libnrniv.a(init.c.o): undefined reference to symbol 'dlclose@@GLIBC_2.2.5'
//usr/lib64/libdl.so.2: error adding symbols: DSO missing from command line

Working on fix locally.

[nrnhines]

This kind of thing seems to be almost entirely experimental and one never knows a priori when a different linker or machine or style (wheel, package, etc.) will need a slightly differen tweak. I don't know whether it would ever be possible or worthwhile to (try to) eliminate certain types of library dependency.