Make failure with Cuda 11.2

User discussions
1

GROMACS version:2020.5
GROMACS modification: No
Here post your question

I’m trying to install Gromacs 2020.5 with cuda 11.2, using gcc 10.2 and openmpi 4.0.5.

After “make”, an error shows up.

[100%] Linking CXX executable …/…/bin/gmx_mpicuda512_2020
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator new(unsigned long)@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to vtable for __cxxabiv1::__vmi_class_type_info@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_begin_catch@CXXABI_1.3' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator delete@GLIBCXX_3.4’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_pure_virtual@CXXABI_1.3' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to typeinfo for std::bad_alloc@GLIBCXX_3.4’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to vtable for __cxxabiv1::__class_type_info@CXXABI_1.3' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to std::nothrow@GLIBCXX_3.4’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator delete(void*, std::nothrow_t const&)@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __gxx_personality_v0@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_guard_acquire@CXXABI_1.3' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_throw@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to std::basic_ostream<char, std::char_traits<char> >& std::__ostream_insert<char, std::char_traits<char> >(std::basic_ostream<char, std::char_traits<char> >&, char const*, long)@GLIBCXX_3.4.9' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to std::ios_base::Init::~Init()@GLIBCXX_3.4’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator new[](unsigned long, std::nothrow_t const&)@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to vtable for __cxxabiv1::__si_class_type_info@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator delete(void*)@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator new(unsigned long, std::nothrow_t const&)@GLIBCXX_3.4’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to operator new[](unsigned long)@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_allocate_exception@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_guard_release@CXXABI_1.3' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_rethrow@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to std::terminate()@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to __cxa_end_catch@CXXABI_1.3’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to std::bad_alloc::~bad_alloc()@GLIBCXX_3.4' /truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to std::ios_base::Init::Init()@GLIBCXX_3.4’
/truba/home/merts/sw/cuda_11.2/lib64/libcufft.so: undefined reference to `vtable for __cxxabiv1::__enum_type_info@CXXABI_1.3’
collect2: error: ld returned 1 exit status
make[2]: *** [bin/gmx_mpicuda512_2020] Error 1
make[1]: *** [src/programs/CMakeFiles/gmx.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs…
[100%] Linking CXX static library …/…/…/…/lib/libgpu_utilstest_cuda.a
[100%] Built target gpu_utilstest_cuda
make: *** [all] Error 2

cmake options used are below:

cmake … -DCMAKE_INSTALL_PREFIX=$HOME/sw/gromacs-2020.5_gcc10.2_openmpi4.1_CUDA_AVX512 -DGMX_SIMD=AVX_512 \

-DGMX_GPU=on -DGMX_MPI=on -DGMX_DEFAULT_SUFFIX=OFF \

-DGMX_BINARY_SUFFIX=_mpicuda512_2020 -DGMX_LIBRARY_SUFFIX=_mpicuda512_2020 \

-DGMX_BUILD_OWN_FFTW=ON -DCUDA_TOOLKIT_ROOT_DIR=$HOME/sw/cuda_11.2/

2

I don’t think it’s directly your problem, but cuda 11.2 doesn’t support gcc 10 (see Installation Guide Linux :: CUDA Toolkit Documentation). Normally you’d have to modify the cuda headers manually to even get this far. How does it go with gcc 9?

3

If you still have problems, please reopen this issue: CUDA 11.2 build issues (#3880) · Issues · GROMACS / GROMACS · GitLab

4

Sorry for my late response! I was able to install CUDA with 11.2 with gcc 9 and also CUDA 10 with gcc 9. And everything works well for now. Currently using cuda 10 with gmx 2020.5