{math}[foss/2022.05] Hypre v2.24.0, METIS v5.1.0, MPFR v4.1.0, ... w/ Python 3.10.4

[jfgrimm]

requires:

• {numlib}[foss/2022.05] HPL v2.3 (incl. FFTW.MPI v3.3.10 + ScaLAPACK v2.2.0) #15561
• {lang}[foss/2022.05] SciPy-bundle v2022.05 w/ Python 3.10.4 #15563
• {devel}[GCC/11.3.0] Boost v1.79.0 #15572

(created using eb --new-pr)

[jfgrimm]

Test report by @jfgrimm
FAILED
Build succeeded for 3 out of 7 (7 easyconfigs in total)
himem06.pri.viking.alces.network - Linux CentOS Linux 7.9.2009, x86_64, Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz (skylake_avx512), Python 3.6.8
See https://gist.github.com/9174f70e03bdc22da65b572881ef6b23 for a full test report.

[jfgrimm]

Test report by @jfgrimm
FAILED
Build succeeded for 5 out of 7 (7 easyconfigs in total)
himem06.pri.viking.alces.network - Linux CentOS Linux 7.9.2009, x86_64, Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz (skylake_avx512), Python 3.6.8
See https://gist.github.com/2a41752f1d8f33a24ac71ca2bfa77393 for a full test report.

[boegel]

Test report by @boegel
FAILED
Build succeeded for 5 out of 7 (7 easyconfigs in total)
node3502.doduo.os - Linux RHEL 8.4, x86_64, AMD EPYC 7552 48-Core Processor (zen2), Python 3.6.8
See https://gist.github.com/44baa5b65b1139d446558adaa615f34c for a full test report.

[boegel]

@jfgrimm With the different failures, and the fact that we should maybe fatten up the PETSc installation a bit (like in #15519), it's probably better to break this apart into smaller PRs?

[satishskamath]

@boegel and @jfgrimm
SuiteSparse build fails and because SuiteSparse fails, PetSc also fails.
Reason:
There is a step to copy this folder GraphBLAS in the SuiteSparse repo during the installation step:

SuiteSparse-5.12.0/GraphBLAS/CUDA/test

. This folder looks like this:

[satishk@tcn1 test]$ ls -la
total 2934
drwxr-sr-x 2 satishk satishk    4096 Apr 10 22:44 .
drwxr-sr-x 6 satishk satishk    4096 Apr 10 22:44 ..
-rw-r--r-- 1 satishk satishk     430 Apr 10 22:44 AxB_dot3_cuda_tests.cpp
-rw-r--r-- 1 satishk satishk     491 Apr 10 22:44 cuda_tests_template.cpp
-rw-r--r-- 1 satishk satishk   12250 Apr 10 22:44 dataFactory.hpp
-rw-r--r-- 1 satishk satishk   10094 Apr 10 22:44 gen_test_data.ipynb
-rw-r--r-- 1 satishk satishk    1007 Apr 10 22:44 GpuTimer.h
lrwxrwxrwx 1 satishk satishk      35 Apr 10 22:44 graphblascuda_test -> ../../build/CUDA/graphblascuda_test
-rw-r--r-- 1 satishk satishk   72666 Apr 10 22:44 jitTestFactory.hpp
-rw-r--r-- 1 satishk satishk    3035 Apr 10 22:44 Makefile
-rw-r--r-- 1 satishk satishk 2014800 Apr 10 22:44 output
-rw-r--r-- 1 satishk satishk  810533 Apr 10 22:44 outputc
-rw-r--r-- 1 satishk satishk     182 Apr 10 22:44 rmm_log.txt
-rw-r--r-- 1 satishk satishk     924 Apr 10 22:44 run_tests.cpp
-rw-r--r-- 1 satishk satishk    4125 Apr 10 22:44 semiringFactory.hpp
-rw-r--r-- 1 satishk satishk    4737 Apr 10 22:44 test_data.hpp
-rw-r--r-- 1 satishk satishk    4865 Apr 10 22:44 testGen_cmake.py
-rw-r--r-- 1 satishk satishk    3198 Apr 10 22:44 testGen.py
-rw-r--r-- 1 satishk satishk    1097 Apr 10 22:44 test_jitify.cpp
-rw-r--r-- 1 satishk satishk     661 Apr 10 22:44 test_utility.hpp

The copy step fails while copying that symbolic link because CUDA folder is not built on a node without GPUs.

[Wahid612]

Hi,
We made it work on our side by adding:
prebuildopts += "unlink %(builddir)s/SuiteSparse-%(version)s/GraphBLAS/CUDA/test/graphblascuda_test && "

[boegel]

It sounds like we'll need to make sure that symlinks are not being resolved when files are being copied during the installation step. Copying a symlink itself should never cause trouble, the issue here is that the symlink is being resolved.

The SuiteSparse easyblock is still using shutil.copytree and shutil.copy2, rather than our own copy_dir and copy_file (available in easybuild.tools.filetools). You can pass a symlinks=True to copy_dir to avoid trouble with broken symlinks, so that should be used in the SuiteSparse easyblock...

[Wahid612]

@boegel That's indeed a better way. I'll test it and let you know. Should I open an issue for that modification?

[satishskamath]

easybuilders/easybuild-easyblocks#2790 Made a PR for the suitesparse easyblock. As suggested, used copy_dir tool with symlinks=True.

[satishskamath]

#16275 There is already SuiteSparse 5.13.0 for foss 2022a. Should the dependency list here be upgraded for PETSc? @boegel and @jfgrimm

[satishskamath]

#16354 easybuilders/easybuild-easyblocks#2796