[CP2K-user] [CP2K:22425] Re: Is it possible to compile Cp2k-2026.2 using AOCC, AOCL, and OpenMPI?

[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 :)

在2026年9月6日星期日 UTC+8 19:02:36<yis... at 163.com> 写道：

> 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.
> 在2026年8月12日星期三 UTC+8 23:34:11<Zilong He> 写道：
>
>> 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 
>> <https://dftd3.readthedocs.io/en/latest/installation.html#cmake-based-build> 
>> 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 
>> <https://github.com/dftd3/simple-dftd3/blame/a25531005ceda936b61fe7ef8f3896eeca1da75f/src/dftd3/api.f90>: 
>> 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.
>> On Monday, July 27, 2026 at 7:55:39 PM UTC+8 Johann Pototschnig wrote:
>>
>>> 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?
>>>
>>>
>>> On Monday, July 27, 2026 at 3:00:32 AM UTC+2 yis... at 163.com wrote:
>>>
>>>> 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
>>>>
>>>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/c6687b9b-2e1f-4be9-83d2-0d8031c9086fn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260907/3f5dd572/attachment.htm>