Is it possible to compile Cp2k-2026.2 using AOCC, AOCL, and OpenMPI?

[yis...@163.com]

To the Cp2k Developers:

Hello,

I am still a beginner with Cp2k. A couple of days ago, I successfully compiled VASP 6.x using AOCC, AOCL, and OpenMPI. In my tests, I found that this compilation combination indeed provides performance benefits on AMD ZEN3 platforms, resulting in faster computation.

Therefore, I would like to ask whether it is also possible to compile Cp2k 2026.2 using AOCC, AOCL, and OpenMPI. Recently, during my compilation attempts, I keep encountering errors at the tblite compilation step. The specific error messages are attached for your reference. Could you please take a look and offer some suggestions on how to resolve this issue?

Thank you very much for your time and help.

                                                             Sincerely,
                                                               yisichi

[Johann Pototschnig]

If you don't need GFN2-xTB tblite can be deactivated. 

Otherwise you could try to install it yourself beforehand to deal with the error. 

Do you know which compilers are used in dftd3?

How do you install CP2K? Via spack or the toolchain?

[Zilong He]

The truncated error message in the .txt points to problems for s-dftd3/src/dftd3/api.f90 from the tblite subproject simple-dftd3. Thus, it is worth following the instructions for a CMake-based build of simple-dftd3 with the compiler to see if the behavior can be reproduced.

I have no experience with AOCC, but there is a wild guess of mine after seeing the source file in the original repository: somehow, the AOCC deemed some ampersands for fortran line breaking/continuing as "syntax error at or near identifier namespace" and could not resolve the name and argument specifications for the subroutines and functions in api.f90. Sucha hypothesis can be tested just by compiling a very tiny fortran program from the Internet. If this is confirmed, create an issue at the simple-dftd3 repository for help.

With these in mind, I think it is unlikely to see actions or patches on the CP2K's (or tblite's) side to address it. Whether or not the CP2K developers would like to track AOCC/AOCL support explicitly with dedicated regression tests is a separate question.

[yis...@163.com]

Thank you for your response. I believe CP2K's computational efficiency is already sufficiently high. The reason I used AOCC and AOCL to compile CP2K was simply to see if it could run even faster. I noticed on GitHub that the developers are already working hard to improve CP2K's functionality, so I won't raise this as a feature request.

[Shiyang Wang (W0728)]

The AOCC/AOCL compilation is harder to be tracked by CP2K Dashboard then Intel compilers, because AMD doesn't provide official Docker images. We can use Spack instead but not the `./make_cp2k.sh` approach.

For convenience of handling with development topic, please continue the discussion on GitHub :)