{toolchain}[pomkl/2016.04] HPL 2.1 (REVIEW)#3044
{toolchain}[pomkl/2016.04] HPL 2.1 (REVIEW)#3044boegel merged 5 commits intoeasybuilders:developfrom
Conversation
|
@bartoldeman, @ocaisa, @damianam, @rjeschmi: please review? |
| # needs to include -ldl or else we get | ||
| # .../libmkl_core.a(mkl_memory_patched.o): undefined reference to symbol 'dlsym@@GLIBC_2.2.5' | ||
| import os | ||
| os.environ['LIBS'] = '-ldl' |
There was a problem hiding this comment.
@bartoldeman I'd like to figure this out and get it fixed in the toolchain support in framework, rather this via this 'hack'...
|
Easyconfigs unit test suite PASSed (see https://jenkins1.ugent.be/job/easybuild-easyconfigs-pr-builder/7999/console for more details). This pull request is now ready for review/testing. Please try and find someone who can tackle this; contact @boegel if you're not sure what to do. |
|
Test report by @boegel |
|
Test report by @boegel |
|
Jenkins: test this please |
|
Test report by @boegel |
|
Test report by @boegel |
|
Test report by @boegel edit: failing test caused by outdated easyblock for Intel MKL, please ignore, will retest with imkl easyblock in develop (required for PGI support, cfr. easybuilders/easybuild-easyblocks#866) |
|
Test report by @boegel |
|
Test report by @boegel |
|
tested on top of easybuilders/easybuild-framework@67abf99 |
|
I'd love to get some feedback on this from @rjeschmi, @bartoldeman, @damianam or @ocaisa before this goes in... |
| source_urls = ['http://www.open-mpi.org/software/hwloc/v%(version_major_minor)s/downloads/'] | ||
| sources = [SOURCE_TAR_GZ] | ||
|
|
||
| # numactl is a direct dependency of PGI |
There was a problem hiding this comment.
Even if numactl is a direct dependency of PGI, I think it doesn't hurt to have it here also (at least if you are using minimal toolchains). In my opinion if we reference $EBROOTSOMETHING this something should be a dependency or an intrinsic part of the toolchain (compiler/mpi/math), specially if all the other versions that package have it as a dependency. Otherwise I see it as error prone (if PGI drops numactl as a dependency for instance).
There was a problem hiding this comment.
Maybe include it as ('numactl', '2.0.11', '', ('GCCcore', gccver))?
There was a problem hiding this comment.
I agree that it would be better to include numactl as an explicit dependency, but then again hardcoding a subtoolchain (which would be required in this case) will mess up things like --try-toolchain.
On the other hands, without including numactl as an explicit dep, you'll run into problems too with --try-toolchain, since you would have to go and dive in and add the numactl dependency for it to work. Hmm. :-)
Since we're screwed either way, I feel including numactl as an explicit dep is better, indeed.
|
Test report by @boegel |
|
last test was on top of easybuilders/easybuild-easyblocks#919 |
|
I can confirm that it works here, and I am fine with the explicit numactl dependency. |
|
Thanks for the feedback @bartoldeman! |
|
Test report by @boegel edit: tested on top of easybuilders/easybuild-framework#1764 |
|
Test report by @boegel |
|
Going in, thanks for the feedback @bartoldeman , @damianam and @rjeschmi! |
PGI-based toolchain for latest PGI 16.4 on top of GCC 5.3.0 and binutils 2.26
requires
easybuilders/easybuild-easyblocks#916andeasybuilders/easybuild-framework#1761