Compilation fails with CUDA and gcc 5
When compiling with CUDA and gcc 5+ compilation fails on libgromacs_generated_nbnxn_cuda.cu.
This problem is known: https://mailman-1.sys.kth.se/pipermail/gromacs.org_gmx-users/2016-May/105608.html.
and it is very confusing for unexperienced users since cmake indicates no problems but the build fails.
Cmake should either emit an error when such combination of nvcc and gcc occurs or the fix indicated in the link should be added.
#2 Updated by Mark Abraham over 3 years ago
Technically the problem lies with using a version of glibc (upon which gcc depends) that no released version of CUDA supports, so a proper action is to use gcc 4.9. However the work-around there is indeed effective.
We could try to backport our workaround from #1982.