Add configure target and sample jobscript for Helma CPU cluster.

[jonasschwab]

Also adds libfakeintel.so to spoof Intel CPUs for MKL. See e.g. https://danieldk.eu/Software/Misc/Intel-MKL-on-AMD-Zen

[Copilot]

Pull request overview

Adds support for building/running ALF on the NHR@FAU Helma CPU cluster and introduces an optional MKL “fake Intel CPU” preload library to improve MKL performance on AMD CPUs.

Changes:

• Add HELMA machine configuration to configure.sh (module setup + IntelLLVM toolchain + MKL/HDF5 wiring).
• Add Scripts_and_Parameters_files/JobfileHelma.sh sample Slurm job script (incl. optional LD_PRELOAD for MKL spoofing).
• Add Libraries/Modules/fakeintel.c + Makefile rule to build libfakeintel.so; also update .gitignore and fix typos in existing job scripts.

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
configure.sh	Adds HELMA machine case to set modules/compiler flags and HDF5 handling.
Scripts_and_Parameters_files/JobfileSuperMUC.sh	Comment typo fixes.
Scripts_and_Parameters_files/JobfileSuperMUC-NG.sh	Comment typo fix.
Scripts_and_Parameters_files/JobfileFritz.sh	Comment typo fix.
Scripts_and_Parameters_files/JobfileHelma.sh	New Helma Slurm job script + optional LD_PRELOAD guidance for MKL spoofing.
Libraries/Modules/fakeintel.c	New C shim exporting MKL “Intel CPU true” symbols for preloading.
Libraries/Modules/Makefile	Builds libfakeintel.so and adds it to the lib target; cleans it.
.gitignore	Ignores generated *.so artifacts.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[johanneshofmann87]

Hi @jonasschwab I just stumbled across this PR, and we might be suffering similar performance issues. Quick questions: As I understand it, one needs to use the preload mechanism, which I couldn't find in the HELMA script. If we want to adapt our local cluster scripts, where do I have to put the preloading? Or is something else taking care of it?

[jonasschwab]

Hi @jonasschwab I just stumbled across this PR, and we might be suffering similar performance issues. Quick questions: As I understand it, one needs to use the preload mechanism, which I couldn't find in the HELMA script. If we want to adapt our local cluster scripts, where do I have to put the preloading? Or is something else taking care of it?

To preload the shared object file, one just needs to set the environment variable LD_PRELOAD pointing to the file. In the Helma sample jobscript, the following line is commented out:

export LD_PRELOAD=/absolute/path/to/ALF/Libraries/Modules/libfakeintel.so

[jonasschwab]

Since I already have you here @johanneshofmann87: Where do you think is the best location to put fakeintel.c / libfakeintel.so? I put it in Libraries/Modules and it gets automatically compiled with all the rest, but conceptually it might better fit into Scripts_and_Parameters_files/.

[johanneshofmann87]

Since I already have you here @johanneshofmann87: Where do you think is the best location to put fakeintel.c / libfakeintel.so? I put it in Libraries/Modules and it gets automatically compiled with all the rest, but conceptually it might better fit into Scripts_and_Parameters_files/.

Hmm, I see your point, but I'm not sure I like Scripts_and_Parameters_files/ better. While it is kind of a runtime file, it is more of (fake) library rather than a script (or parameter file). W could keep it in Library, but move it one level up? Not sure if this is sooo much better...

[fassaad]

Helma is down for the next weeks, so that we will wait till the machine is up again to test and then merge.

[fassaad]

The group here has been using this, and it works just fine. Hence I think that we could merge.

[fassaad]

All good.