Warning during compilation on Ubuntu using GNU tools
Posted: Tue Jul 06, 2021 12:53 pm
Using the guide at wiki/index.php/A_Ubuntu_based_installation_of_VASP, VASP compiles without issues on Ubuntu 18.04 LTS, but using the exact same procedure on Ubuntu 20.04 LTS produces a warning during compilation (for each binary and libdmy.a):
This warning is reproducible on both Intel and AMD cpus. Intel's compiler and libraries work on Intel-based Ubuntu 20.04 systems without any issues. However, we are going to invest into a set of workstations equipped with AMD running Ubuntu 20, as they offer many more CPU cores at a lower price (and most of our researchers at the department do not perform VASP computations, or anything else that might perform better on an Intel-based system). Our initial tests show that VASP compiled on AMD using open-source libraries and gfortran runs about as fast as VASP compiled on a similar Intel CPU, using Intel's oneAPI suite (Intel fortran, Intel MKL, Intel MPI). The models even converged to very similar energies. However, the one compiled with open-source tools produced the following warning after each VASP run (only to console. Not to any VASP output file):
I noticed that Ubuntu 18.04 comes with GNU Fortran (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0, while Ubuntu 20.04 ships with GNU Fortran (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0. I tried installing gfortran 7.5 on Ubuntu 20 and the warning above, concerning source file drdatab.f90, vanished, but it opened up a whole new can of worms with a lot of other errors (not just warnings) that made the compilation stop.
Would it be possible for the support team to come up with an installation guide that works with Ubuntu 20.04 and open-source tools?
Code: Select all
gcc -E -P -C -w drdatab.F >drdatab.f90
mpif90 -O1 -ffree-form -ffree-line-length-none -c -o drdatab.o drdatab.f90
drdatab.f90:212:0:
212 | BUFLIN=BUFLIN(2:LMAX)
|
Warning: ‘__builtin_memmove’ reading 32767 bytes from a region of size 32766 [-Wstringop-overflow=]
Code: Select all
Note: The following floating-point exceptions are signalling: IEEE_UNDERFLOW_FLAG IEEE_DENORMAL
Would it be possible for the support team to come up with an installation guide that works with Ubuntu 20.04 and open-source tools?