add patch for FFTW 3.3.8 to avoid use of -no-gcc when building with Intel compilers, to fix installation on CentOS 8 (cfr. #10932)

[boegel]

(created using eb --new-pr)

cfr. #10932

[boegel]

@boegelbot please test @ generoso

[boegelbot]

@boegel: Request for testing this PR well received on generoso

PR test command 'EB_PR=11050 EB_ARGS= sbatch --job-name test_PR_11050 ~/boegelbot/eb_from_pr_upload_generoso.sh' executed!

• exit code: 0
• output:

Submitted batch job 4697

Test results coming soon (I hope)...

Details

- notification for comment with ID 667662833 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

[boegelbot]

Test report by @boegelbot
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in this PR)
generoso-x-5 - Linux centos linux 8.2.2004, x86_64, Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz (haswell), Python 3.6.8
See https://gist.github.com/6a87ac4867615eb895e1b9aabcaad8bc for a full test report.

[boegel]

Test report by @boegel
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in this PR)
node3401.kirlia.os - Linux centos linux 7.8.2003, x86_64, Intel(R) Xeon(R) Gold 6240 CPU @ 2.60GHz (cascadelake), Python 2.7.5
See https://gist.github.com/453cc08c054917e6c3e52f58f73cfc6c for a full test report.

[boegel]

Test report by @boegel
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in this PR)
node3116.skitty.os - Linux centos linux 7.8.2003, x86_64, Intel(R) Xeon(R) Gold 6140 CPU @ 2.30GHz (skylake_avx512), Python 3.6.8
See https://gist.github.com/c3aa7921de9f0cebf5254d1709aae337 for a full test report.

[boegel]

Test report by @boegel
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in this PR)
node2402.golett.os - Linux centos linux 7.8.2003, x86_64, Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz (haswell), Python 2.7.5
See https://gist.github.com/dc6983bbfb0809f01b5fb921978e01bc for a full test report.

[casparvl]

The patch looks fine to me. Also, at IDRIS, they already removed the -no-gcc some time ago and have no issues (according to FFTW/fftw3#184), so that seems to support this approach.

One question: on the same issue, it was mentioned that failures occured regardless of Intel Compiler version, as long as the GCC in the path was <= 6.5.0. There are a number of EasyConfigs that use FFTW-3.3.8 with the 2018b toolchain generation. Did you try those on CentOS 8.2? Do they work?

And, does this also apply to older FFTW versions? I'm not saying we should backport everything, everywhere. But maybe we can log it in your issue (along with the ones we DID patch, for clarity) #10932

[boegel]

One question: on the same issue, it was mentioned that failures occured regardless of Intel Compiler version, as long as the GCC in the path was <= 6.5.0. There are a number of EasyConfigs that use FFTW-3.3.8 with the 2018b toolchain generation. Did you try those on CentOS 8.2? Do they work?

@casparvl I didn't test with intel/2018b, because it seems that toolchain is too old to use on CentOS 8 (it trips over changes in glibc). Not planning to spend more time on that, not worth the trouble imho...

And, does this also apply to older FFTW versions? I'm not saying we should backport everything, everywhere. But maybe we can log it in your issue (along with the ones we DID patch, for clarity) #10932

It does apply to older FFTW versions, but I'm not sure if it's worth the trouble to try and fix the installation of older FFTW versions too?

[casparvl]

I can imagine why you called it a 'dirty' patch, but since EasyBuild officially has GNU/linux systems as a requirement, I would say it's fine.

I would personally prefer to also add FFTW-3.3.8-intel-2018b.eb in here. We might as well cover all FFTW-3.3.8's, and this would be the oldest. (I might have some personal bias since we still have a pretty extensive 2018b stack ;-))

Ok scratch that, just read your comment above. You're right, intel-2018b would run into glibc problems anyway.

[casparvl]

but I'm not sure if it's worth the trouble to try and fix the installation of older FFTW versions too?

No, I agree. And with your argument that glibc also breaks for those older toolchains, there's no use fixing the FFTWs anyway.

[casparvl]

Test report by @casparvl
FAILED
Build succeeded for 0 out of 1 (1 easyconfigs in this PR)
tcn900.bullx - Linux RHEL 7.7, x86_64, Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz, Python 2.7.5
See https://gist.github.com/45d4a862974ad5acd657e13070d81a2e for a full test report.

[casparvl]

Test report by @casparvl
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in this PR)
software1.lisa.surfsara.nl - Linux debian 10.5, x86_64, Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz, Python 2.7.16
See https://gist.github.com/c6ae0ac8a8646ab554a60acd61051b82 for a full test report.

[casparvl]

Test report by @casparvl
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in this PR)
tcn900.bullx - Linux RHEL 7.7, x86_64, Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz, Python 2.7.5
See https://gist.github.com/4c727e68cbb02fc3d56e4b7f27125c87 for a full test report.

[casparvl]

Failing test was because of the fact that FFTW doesn't like an optarch that doesn't match the AVX capability of the host. That has been a (minor) issue for a long time; not relevant in the context of this PR. Time to merge! Thanks @boegel

[klust]

The patch does not only solve problems on CentOS 8. Compiling FFTW 3.3.8 with Intel 2020 update 2 also failed on CentOS 7.8 (where update 1 still worked) but at a different point then on CentOS 8. However, this patch also solved that problem.