Open MES-physics opened 1 year ago
I have never seen this issue before, even though we do routinely use the soap_turbo
library linked with OpenBLAS. This seems to be related to threading support in your OpenBLAS installation. What happens if you do export OMP_NUM_THREADS=1
before running QUIP? It could be due to some interoperability issue between QUIP and soap_turbo. How did you compile QUIP (e.g., which makefile did you use)?
Do you have a fresh installation of QUIP+GAP+soap_turbo? I.e., did you do git clone --recursive ...
or have you pulled the repo? If you pulled the QUIP repo and forgot to update the soap_turbo submodule perhaps you are running an old version of the library.
Any ideas, @janklinux?
Yes, I have seen this before though it never crashed for me. It is an issue of how the openblas was compiled, as it complains. At compile time one needs to set USE_OPENMP=1 to make it openmp aware, that does not mean it uses threads, this is specified with the OMP_NUM_THREADS=N at runtime.
I got it by git clone --recursive
. It's the same one I've been using for some months with regular soap.
I'm thinking this is related to soap_turbo because it's the first time I've seen this problem. This is using the soap_turbo potential obtained from a serial run.