Questions regarding the compilation of VASP on various platforms: hardware, compilers and libraries, etc.
Moderators: Global Moderator, Moderator
-
Andy.J
- Newbie
- Posts: 2
- Joined: Tue Jul 30, 2024 9:16 pm
#1
Post
by Andy.J » Thu Aug 01, 2024 6:15 pm
Hi,
I'm trying to compile the vasp 6.4.3 within the NVIDIA HPC-SDK version 23.5 ubuntu 20.04 docker container. The makefile.include is based on the makefile.include.nvhpc_omp_acc. I have compiled the fftw-3.3.10 within the docker container with command
Code: Select all
./configure --enable-shared --enable-threads --enable-openmp --prefix=/install-path
and
. The nvcc --version shows the following 'Cuda compilation tools, release 12.1, V12.1.105'. The mpirun --version shows the following 'mpirun (Open MPI) 4.1.5rc2'. Could you help me figure out the reason for the error? Thanks.
Andy
You do not have the required permissions to view the files attached to this post.
-
manuel_engel1
- Global Moderator
- Posts: 125
- Joined: Mon May 08, 2023 4:08 pm
#2
Post
by manuel_engel1 » Fri Aug 02, 2024 7:45 am
Hi Andy,
Thanks for posting on the VASP forum. The error message you get is related to
cuSOLVERMp. I think there is a version conflict with this library. When you look into the makefile.include, you will find the following block:
Code: Select all
# Use cusolvermp (optional)
# supported as of NVHPC-SDK 24.1 (and needs CUDA-11.8)
CPP_OPTIONS+= -DCUSOLVERMP -DCUBLASMP
LLIBS += -cudalib=cusolvermp,cublasmp -lnvhpcwrapcal
The version of NVHPC-SDK you are using to compile VASP is 23.5, but we require at least version 24.1 for this feature to work. I recommend that you update the nvidia SDK to at least version 24.1 and try again. Alternatively, you can disable this feature by commenting out the two lines in the snippet above.
Manuel
VASP developer
-
Andy.J
- Newbie
- Posts: 2
- Joined: Tue Jul 30, 2024 9:16 pm
#3
Post
by Andy.J » Fri Aug 02, 2024 6:23 pm
That solution works for me. Thank you very much!
Andy