Possible Issue with Arch Files

[Rashid Riboul]

Hi, I'm trying to install CP2K using the designated Arch file and I saw an interesting message:

==================== Getting proc arch info using OpenBLAS tools ====================
wget  --quiet https://www.cp2k.org/static/downloads/OpenBLAS-0.3.27.tar.gz
OpenBLAS-0.3.27.tar.gz: OK
Checksum of OpenBLAS-0.3.27.tar.gz Ok
cc1: error: unknown value 'apple-m1' for '-march'
cc1: note: valid arguments are: armv8-a armv8.1-a armv8.2-a armv8.3-a armv8.4-a armv8.5-a armv8.6-a armv8.7-a armv8.8-a armv8-r armv9-a armv9.1-a armv9.2-a armv9.3-a native
cc1: note: did you mean '-mcpu=apple-m1'?
C Compiler (/opt/homebrew/bin/gcc) is something wrong.
cc1: error: unknown value 'apple-m1' for '-march'
cc1: note: valid arguments are: armv8-a armv8.1-a armv8.2-a armv8.3-a armv8.4-a armv8.5-a armv8.6-a armv8.7-a armv8.8-a armv8-r armv9-a armv9.1-a armv9.2-a armv9.3-a native
cc1: note: did you mean '-mcpu=apple-m1'?
C Compiler (/opt/homebrew/bin/gcc) is something wrong.
cc1: error: unknown value 'apple-m1' for '-march'
cc1: note: valid arguments are: armv8-a armv8.1-a armv8.2-a armv8.3-a armv8.4-a armv8.5-a armv8.6-a armv8.7-a armv8.8-a armv8-r armv9-a armv9.1-a armv9.2-a armv9.3-a native
cc1: note: did you mean '-mcpu=apple-m1'?
C Compiler (/opt/homebrew/bin/gcc) is something wrong.
make: *** [config.h] Error 1
Makefile.system:1546: Makefile.: No such file or directory
make: *** No rule to make target `Makefile.'.  Stop.
ERROR: (/Users/caracallynx/MD/cp2k/tools/toolchain/scripts/get_openblas_arch.sh, line 52) Non-zero exit code detected.
ERROR: (./scripts/stage0/setup_buildtools.sh, line 67) Non-zero exit code detected.

For the record, this is the latest version history of CP2K (downloaded from repository two days ago) and I'm attempting to install it on an Apple Silicon Macbook M1 Max running Sonoma 14.5.

[Krack Matthias]

Hi

 

Thanks for reporting the issue. It should be fixed now by PR #3442 in the CP2K development version.

 

Best

 

Matthias

 

--
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+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/4c2de455-efb6-4886-8f1d-e1c1b2eac4a0n%40googlegroups.com.

[Rashid Riboul]

Thanks Matthias; I downloaded the developmental version, and it compiled fine! The only thing now is that it seems to be failing over 50 regtests. Would that be due to me using the developmental version? Or is something else possibly the issue?

[Krack Matthias]

Did you increase the OMP_STACKSIZE to 16M or more, e.g. with export OMP_STACKSIZE=16M ?

 

HTH

 

Matthias

 

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/9e7679a2-0155-4ed9-8c52-adc07f58af90n%40googlegroups.com.

[Rashid Riboul]

I did, AND increased the ulimit size to 65 MB as well, but still got failures. I posted a pic of the summary after the test. From what I tested, the 4 wrong tests seem to occur if I use the Accelerate Framework as my BLAS/LAPACK libraries. If I use OPENBLAS, the 4 wrong tests don't occur, but I still get the same amount of failures.

Please let me know what other information I can give to help troubleshoot this issue.

[Krack Matthias]

The CP2K Apple M1 tester for the current CP2K development version shows no issues in the log file.

Did you follow all the instructions in the howto, like to deactivate brew’s clang based openmpi installation with “brew unlink open-mpi” to avoid any interference with the openmpi installation which is performed by the CP2K toolchain using the GNU compiler?

 

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/d3d0d560-59dc-490f-aa03-9dcb8e2481ffn%40googlegroups.com.

[Rashid Riboul]

Ah; that fixed most of the failures actually. I still have 1 failed test and 8 wrong tests. I am attempting to install the SSMP version instead of the PSMP version. Would that be better? I figured that the PSMP version wouldn't be necessary for my case, but only because I don't plan on using this machine as an MPI node.

[Krack Matthias]

By contrast to the Apple M1 psmp regression test, the corresponding ssmp test is not run continuously. I have synced the Apple M1 ssmp arch file with the psmp one. An ssmp test run with that file showed no issues on my Apple M1 using 2 OpenMP threads.

 

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/835a2e55-f240-4b21-8ad4-62fcaf29dcdbn%40googlegroups.com.